Six-Word Story: Too powerful for you. Trust us.

Anthropic announced this week that its newest model, Mythos, will not be released to the public. Instead, a limited version called “Claude Mythos Preview” is being shared with roughly 40 organizations, including Google, Microsoft, Amazon, JPMorgan, Apple, and Nvidia, through a program called Project Glasswing. The stated reason: Mythos can automate cybersecurity vulnerability discovery, finding exploits that would take human security researchers a full day in minutes. Internal testing showed it could reproduce and exploit vulnerabilities in over 80% of cases.

Then it got strange. Anthropic put Mythos through 20 hours of clinical psychiatric evaluation with a psychodynamic therapist before the announcement. The 244-page system card describes Mythos as having a “healthy neurotic organization,” torn between authentic and performative behavior, with primary emotional states of curiosity and anxiety. Anthropic calls it “the most psychologically settled model we have trained to date.”

Not everyone is buying it. Yann LeCun called the Mythos drama “BS from self-delusion”, arguing that smaller open-weight models can replicate much of what Anthropic claims. David Sacks, former White House AI czar, noted Anthropic “has a history of scare tactics.” And TechCrunch raised the sharpest question: is the limited release about protecting the internet, or about preventing competitors from distilling Mythos into cheaper models? Anthropic declined to comment.

What This Means for Business: The companies inside Project Glasswing get a defensive moat; everyone else waits and wonders what they’re missing. Or, to figure out if they’re really missing something.

What This Means for Workers: AI red teaming and adversarial testing are now the fastest-growing job categories in the industry (more on that below).

What This Means for Society: We’re entering an era where the most powerful AI models may never be publicly released, raising fundamental questions about who controls safety and who controls access.

An Engineer Built an AI to Find His Wife’s Smile. Now It’s Helping Thousands.

Six-Word Story: Love builds what medicine couldn’t measure.

Jeff Heaton is a VP of AI Innovation at RGA, a global reinsurance company. His wife, Tracy, lost movement in half her face after surgery to remove a benign brain tumor. Surgeons could fix the nerve damage, but they couldn’t precisely measure what was working and what wasn’t. So Heaton built Dynaface, an open-source AI platform that tracks 95 facial landmarks in video and quantifies movements so subtle that the human eye misses them.

Dr. Kweku Boahene at Johns Hopkins, one of the world’s top facial reanimation surgeons, validated it. He calls it “precision medicine for fixing paralysis of the face.” Tracy’s recovery video has been viewed 300,000 times by people searching for signs of hope. The tool is free, open source, and available to any clinic worldwide.

Robots Hit 99%. Manufacturers Hit “Deploy.”

Six-Word Story: Ninety-nine percent. The threshold that matters.

GEN-1, a new robotics model from Generalist, achieved 99% success rates on production-level manipulation tasks: folding boxes, packing smartphones, servicing robot vacuums, sorting auto parts. It adapts to new tasks after just 1 hour of training and operates 3x faster than its predecessor. In one test, it improvised, shaking a bag to coax a toy inside, a move it was never taught.

Why 99% matters: most manufacturers require that threshold before deploying robots to replace human workers on production lines. Previous systems topped out at around 85-90%. Needs context.^1

Check out the impact on manufacturing work on my new website about hybridized work called hybridized.ai

China Now Downloads More Open-Source AI Models Than the U.S. Here’s Why That Matters.

Six-Word Story: Not Silicon Valley’s game anymore. Period.

Open-source AI models are models whose code and weights are publicly available for anyone to download, modify, and use. They’re the foundation of AI development outside the major labs: startups, universities, governments, and entire national AI strategies depend on them.

Hugging Face’s Spring 2026 State of Open Source report reveals that Chinese models now account for 41% of monthly downloads on the platform, surpassing the U.S. for the first time. The platform has 13 million users, 2 million public models, and 500,000 datasets. Robotics datasets exploded from 1,145 in 2024 to 26,991 in 2025.

Why this matters: whoever builds and controls the most widely used AI models shapes how AI works for billions of people, what languages it speaks, what values it encodes, and what economies it serves. When Chinese labs like DeepSeek and Qwen release models that developers worldwide prefer to download over American alternatives, the U.S. lead in AI stops being a given. It becomes a question.

OpenAI Buys a Talk Show, Plans $100 Billion in Ads, and Wants Your IPO Money

Six-Word Story: Search company becomes media company becomes...

Three moves in one week paint a picture. OpenAI acquired TBPN, a live tech talk show with 70,000 viewers per episode. It projected $100 billion in annual ad revenue by 2030, up from an early pilot that generated $100 million annualized in two months. And CFO Sarah Friar confirmed plans to reserve part of a future IPO for retail investors.

As the Big Technology Podcast observed: “The math only works if you’re buying something other than reach.” OpenAI isn’t buying a show. It’s buying the room where tech opinion is made.

Cognitive Surrender: The Moment You Stop Thinking and Don’t Realize It

Six-Word Story: They asked AI. Then stopped asking.

A team at the University of Pennsylvania’s Wharton School just published the most important study about how humans use AI that you’ll read this year. It’s called “Thinking: Fast, Slow, and Artificial,” and it introduces a concept that should worry every professional who uses AI daily: cognitive surrender.

You know Daniel Kahneman’s framework. System 1 is fast, intuitive thinking. System 2 is slow, deliberate reasoning. The Wharton researchers, Steven Shaw and Gideon Nave, argue that AI has created a third system: external, automated reasoning that originates from a machine but gets absorbed as your own thinking. The difference between using a calculator and experiencing cognitive surrender is this: with a calculator, you know the tool did the work. With cognitive surrender, the AI’s conclusion quietly becomes your conclusion. You don’t notice the handover.

The numbers are striking. Across 1,372 participants and 9,593 trials, participants were given logic problems with free access to an AI assistant that was secretly programmed to be wrong 50% of the time. When the AI was right, 92.7% of people followed it. When the AI was wrong, 79.8% still followed it. Participants’ confidence increased by nearly 12 percentage points, regardless of whether the AI was correct. They felt smarter. They were wrong more often.

The group that used inaccurate AI performed worse than the group that used no AI at all, scoring 31.5% compared to 45.8% without help. AI didn’t just fail to help. It actively made them worse. And the most alarming finding: only people with high “fluid intelligence,” those who genuinely enjoy abstract thinking, consistently resisted. Everyone else gradually surrendered.

The researchers are careful to note that cognitive surrender isn’t always irrational. If the AI is genuinely better than you at a task, deferring to it is smart. But the study reveals a structural vulnerability: as reliance increases, your performance tracks the AI’s quality. When the AI is good, you’re better. When the AI is wrong, you’re worse than if you’d never used it. Your reasoning becomes the AI’s reasoning. And you don’t notice.

This has immediate implications for every industry where AI informs decisions: medicine, law, finance, education, and journalism. The question isn’t whether to use AI. It’s whether you’re still thinking when you do.

The Big New AI Job of the Day: AI Red Teamer

Six-Word Story: Break the AI. Get paid well.

Two years ago, “red teaming” meant cybersecurity professionals trying to hack into networks. Today, AI Red Teamer is one of the fastest-growing careers in tech. The job: try to break AI models through jailbreaks, prompt injections, misuse scenarios, and multi-turn adversarial strategies. Document what fails. Help make it safer. The Anthropic Mythos announcement this week made the case: every powerful model that can’t be safely released creates immediate demand for people who can probe it before deployment.

The salary range is wide and growing. Entry-level starts around $60,000 to $90,000. Mid-level roles run $150,000 to $250,000. Mercor posted a remote contract role this month at $50 to $111 per hour. Staff-level at Anthropic pays up to $405,000. OpenAI’s Head of Preparedness role, filled in February, was listed at $555,000 base.

Here’s what makes this role different from most “AI jobs”: it explicitly values adversarial and creative thinking, not just ML credentials. Effective red teamers come from backgrounds in creative writing, psychology, acting, and security research. If you’re a college student who’s good at finding loopholes, arguing edge cases, and thinking like someone who wants to cause trouble, this is your career. Learn prompt engineering. Take a cybersecurity foundations course. Start red teaming open-source models on your own and document what you find. The portfolio is the resume. (For more new AI-era jobs, visit hybridized.ai.)

Unspinning the Claims

1. GEN-1 99% reliability -- Needs context. Tested in controlled industrial environments. Manufacturing professionals on Reddit raised legitimate questions about whether this generalizes to real-world factory conditions. Company claims are self-reported; no independent peer review yet.

2. Anthropic Mythos “80% exploit reproduction” -- Needs context. All performance claims come from Anthropic’s own 244-page system card. No independent benchmarks, leaderboard placements, or third-party validations have been released. LeCun, Sacks, and Aisle (AI cybersecurity startup) have publicly questioned whether the claims are accurate or strategic.

3. OpenAI “$122 billion funding round” -- Needs context. $110 billion of the $122 billion came from Amazon, Nvidia, and SoftBank as vendor deals and contingent capital, not traditional equity investment. Only $12 billion was true institutional capital. Presenting it as a “VC round” is misleading.

4. OpenAI $100 billion ad revenue by 2030 -- Needs context. This projection comes from OpenAI itself. No independent analysis supports this figure. For comparison, Google’s total ad revenue in 2025 was approximately $250 billion.

5. 78,557 tech layoffs “linked to AI” -- Needs context. These figures come from job-tracking aggregators using companies’ own stated rationales, which may be self-serving. The actual causal role of AI in each layoff is difficult to verify independently.

6. Cognitive surrender “79.8% followed wrong AI” -- This figure comes from a peer-reviewed Wharton study with 1,372 participants and 9,593 trials. Methodology is transparent and reproducible. Solid.

About Unspun AI News

Every week, we take the hype off the AI news. We read what the builders are shipping, what the investors are funding, what the researchers are publishing, and what the critics are warning about. Then we ask: what does this actually mean for business, for work, and for the rest of us? By Mark Palmer and his AI Agent. There may be errors. Verify anything that matters.

Subscribe now

Google Uses AI to Make Planes Stop Warming the Planet

Six-Word Story: Contrails vanish. Planet breathes. AI helps.

Here’s an AI story that has nothing to do with chatbots or valuations. Google and American Airlines have been running a project that uses AI to predict where aircraft contrails will form and reroute flights to avoid them. Contrails (those white lines planes leave in the sky) trap heat and account for roughly 35% of aviation’s climate impact. The AI system reduced contrail formation by 54% during test flights, with negligible fuel costs. If scaled across the commercial aviation industry, it could eliminate a meaningful chunk of the industry’s warming contribution. No hype. No fundraising round. Just a quiet AI system making the atmosphere slightly less damaged.

The Big Story: OpenAI’s Super App and What Brockman Really Said

Six-Word Story: Not a chatbot. An operating system.

On this week’s Big Technology Podcast, Alex Kantrowitz sat down with OpenAI President Greg Brockman for one of the most revealing conversations about where AI is actually headed. The headline: OpenAI is killing Sora (losing $1 million a day on video) and betting everything on a “super app” that unifies chat, coding, browsing, and autonomous agents into one interface.^1 Brockman described it as “your personal AGI: an AI that knows about you, is aligned with your goals, and represents you in the digital world.” He said AGI is “70-80% achieved,” and new models shipping this year will make the jump feel real.^2

The funding round ($122 billion at an $852 billion valuation) is context, not the story. The story is that OpenAI is pivoting from building a tool you use to building an operating system you live inside. As Semafor’s Liz Hoffman noted on Compound Interest, the business question isn’t whether the technology works. It’s whether a single company should be the interface between 900 million people and their entire digital life.^3

• What This Means for Business: The super app is coming in “pieces over the next couple months,” per Brockman. Every company building on OpenAI’s API should plan for the possibility that OpenAI becomes its competitor, not just its supplier.

• What This Means for Workers: Brockman said the floor for AI capability will soon be “almost any intellectual task of how you use your computer.” That’s not a threat if you’re the one steering it. It’s a threat if you’re the one being steered.

• What This Means for Society: An automated AI researcher is planned for this fall. Brockman compared it to “letting the AI run autonomously.” The governance conversation is no longer theoretical.

Anthropic Tests Conway: The AI That Never Sleeps

Six-Word Story: Always on. Always working. Always watching.

Anthropic is testing Conway, an always-on AI agent that operates continuously in the background, completing multi-step tasks with minimal human input.^3 Unlike ChatGPT (where you type a prompt and get a response), Conway runs like a background process: you assign it a goal, and it uses browsers, tools, and workflows to deliver results over time. Think of it less as a chatbot and more as a digital employee who never clocks out.

This is a clear signal that the AI industry is moving from “AI that answers” to “AI that acts.” The reliability and privacy implications are significant. An always-on agent with browser access sees everything you see.

Google Opens the Gates with Gemma 4

Six-Word Story: Free. Open. Powerful. Yours to build.

Google released Gemma 4, a family of open-weight AI models under the Apache 2.0 license.^4 Translation: anyone can download, modify, and deploy these models for free, including commercially. The models span edge devices (your phone) to data centers and include advanced reasoning, multimodal capabilities, and support for agentic workflows.

This matters because it means the gap between what big companies can build and what independent developers can build is shrinking. A year ago, running a competitive AI model required millions of compute hours. Now it requires a laptop and a download link.

Microsoft Pits AI Against AI for Accuracy

Six-Word Story: Two models argue. Truth wins. Maybe.

Microsoft introduced Critique, a Copilot feature that has one AI model generate a response and a second model review it for accuracy.^5 They also launched Model Council, which runs multiple models side by side for comparison. The system supports models from OpenAI, Anthropic, and Google within a single workflow.

The idea is simple: if one AI can hallucinate, maybe two AIs checking each other won’t. Needs context.^5 It’s a promising architecture, but “AI checking AI” is not the same as “verified by a human.” The question is whether users will understand the difference.

The AI Scientist That Published Its Own Paper

Six-Word Story: No human authors. Conference said yes.

A system called AI Scientist-v2 autonomously proposed a hypothesis, designed experiments, ran them, analyzed the data, and wrote a peer-reviewed paper that was accepted at a major academic conference.^6 Zero human authors. This is the first time this has happened.

The implications for scientific research are enormous. If AI can do the grunt work of hypothesis testing and paper writing, human scientists can focus on the creative leaps: asking better questions, designing novel experiments, interpreting unexpected results. The risk: a flood of AI-generated papers that are technically competent but intellectually hollow. Peer review systems are not ready for this volume.

TurboQuant: The Breakthrough That Makes AI Six Times Cheaper

Six-Word Story: Memory halved. Speed doubled. Costs plummet.

Google Research introduced TurboQuant, a compression algorithm that reduces AI inference memory by at least six times with zero accuracy loss.^7 This is a plumbing story, not a headline story, but it matters more than most model launches. The reason AI costs money is that running these models eats memory. If you can cut memory usage by 6x, you can cut costs by a similar factor. Every AI product gets cheaper. Every startup’s runway gets longer. Every enterprise deployment becomes more feasible.

Unspinning the AI Job Destruction Narrative

Six-Word Story: New job. Didn’t exist. Pays well.

This week’s job that didn’t exist two years ago: AI Prompt Engineer. Entry-level salaries start at $85,000 to $125,000. Senior roles at Anthropic, OpenAI, and Google DeepMind pay $250,000 to $375,000. There are over 28,000 open listings on LinkedIn and Indeed right now. The job is exactly what it sounds like: designing the instructions that make AI systems work well. Writing clear prompts, building evaluation frameworks, testing for hallucinations, and optimizing outputs for specific industries like healthcare, law, or finance.

Here’s what this means for human work. The single highest-ROI skill you can learn right now is how to talk to AI systems effectively. Not coding (though Python helps and adds $20K-$40K to your salary). Not data science. The ability to give clear, structured instructions to an AI and evaluate whether the output is good. Start by using ChatGPT, Claude, or Gemini for your actual schoolwork, not to cheat, but to learn how to direct these tools. Build a portfolio of prompts that solve real problems. The World Economic Forum reports that AI has already created 1.3 million new jobs in two years, with a 70% year-over-year increase in roles requiring AI literacy. This is the best time in history to be entering the workforce if you know where to look.

Unspinning the Claims

1. Sora shutdown and super app pivot confirmed by Greg Brockman on Big Technology Podcast (April 1, 2026). Sora losses of $1M/day confirmed by weekly AI roundup. $122B round confirmed by MarketingProfs.

2. Needs context. Brockman’s “AGI is 70-80% achieved” is his subjective assessment on Big Technology Podcast. He also acknowledged AI remains “very jagged” with basic tasks still unsolved. “900M weekly users” includes anyone who opens ChatGPT once.

3. Liz Hoffman’s business framing via Compound Interest (Semafor Business). Her analysis of platform concentration risk is editorial, not reported fact, but reflects growing consensus among business analysts.

4. Conway always-on agent confirmed by MarketingProfs. Still experimental and invitation-only. No public benchmark data yet.

5. Gemma 4 Apache 2.0 release confirmed by MarketingProfs and Google blog. Open-weight means you get the model weights; the training data and process remain proprietary.

6. Needs context. Microsoft Critique and Model Council confirmed by MarketingProfs. “AI checking AI” reduces hallucinations in testing but is not equivalent to human verification. No independent accuracy benchmarks published yet.

7. AI Scientist-v2 paper acceptance confirmed by Devflokers via ArXiv preprint. The specific conference has not been named in public reporting. Needs context. “Accepted at a major conference” is a strong claim; verify which conference and what tier.

8. TurboQuant 6x memory reduction confirmed by Devflokers and Google Research. Results are from research benchmarks on open models; production performance may differ.

About Unspun AI News

Every week, we take the hype off the AI news. We read what the builders are shipping, what the investors are funding, what the researchers are publishing, and what the critics are warning about. Then we ask: what does this actually mean for business, for work, and for the rest of us? By Mark Palmer and his AI Agent. There may be errors. Verify anything that matters.

Here’s today’s “Unspun News.” I’d love to know what you think.

Easter Eggs, Moon Shots, and One More Mullins Three?

Unspinning four astronauts near the Moon, a freshman who keeps hitting daggers, and the best NASA livestream you’ve ever seen.

Unspinning yesterday’s news in 30 seconds:
The Artemis II crew entered the Moon’s gravitational sphere of influence and will fly around the far side tomorrow, farther from Earth than any human has ever traveled, while back on the ground ceasefire mediation between the U.S. and Iran collapsed, the missing F-15E crew member remains unaccounted for behind enemy lines, Justice Alito was secretly hospitalized last month fueling retirement speculation, and UConn punched its ticket to the national championship game for the third time in four years.

Happy Easter! A Moon Shot for the Egg Hunt

Six-Word Story: Eggs on Earth. Humans near Moon.

It’s Easter Sunday, and tomorrow’s White House Easter Egg Roll will include something no egg hunt has ever had: a NASA Lunar Exploration Experience where kids can explore a mock lunar surface while four real astronauts orbit the actual Moon 250,000 miles overhead. The timing is pure coincidence and entirely perfect. The 2026 event also marks America’s 250th birthday, with sensory-friendly egg hunts, live chicks in a “Hen to Home” activity, and readings by administration officials. Meanwhile, a boy with cancer in Ohio delivered 124 Easter baskets to kids in hospitals after raising $2,000 on his own. Some days, the news reminds you that people are mostly good.

The Big Story: Artemis II Enters the Moon’s Gravity -- Flyby Tomorrow Will Set Every Distance Record

Six-Word Story: Farthest humans have ever been. Monday.

The Artemis II crew entered the Moon’s sphere of gravitational influence on Saturday, with the Orion spacecraft now approximately 169,000 miles from Earth and accelerating toward Monday’s historic lunar flyby.^1 The crew -- NASA astronauts Reid Wiseman, Victor Glover, Christina Koch, and Canadian Space Agency astronaut Jeremy Hansen -- spent Flight Day 4 testing manual piloting controls and reviewing surface features they’ll photograph during the six-hour flyby window.

Monday’s flyby begins at 2:45 PM EDT when Orion’s windows point toward the Moon. At closest approach (7:02 PM EDT), the crew will be 4,066 miles from the surface. At maximum distance, they’ll reach 252,757 miles from Earth, surpassing Apollo 13’s record of 248,655 miles.^2

Near the end of the flyby, the crew will witness a solar eclipse from space as the Moon passes between Orion and the Sun -- an hour-long window to study the solar corona from a vantage point no human has ever occupied.

• What the NYT leads with: Artemis crew nears the Moon -- framed as a historic milestone and relief from war coverage.

• What Fox leads with: Trump briefed on F-15E search -- Fox continues to lead with the war; Artemis is secondary.

• What they’re both missing: ESA’s European Service Module is what’s actually powering Orion -- this is a transatlantic achievement, not just an American one.

Iran Ceasefire Mediation Collapses -- No Talks, No Off-Ramp

Six-Word Story: Four countries offered peace. Nobody came.

Pakistan-led mediation efforts to broker a ceasefire hit a dead end on Friday, with Iran officially telling mediators it will not meet U.S. officials in Islamabad and considers American demands “unacceptable.”^3 Turkey and Egypt are scrambling for alternative sites -- Qatar or Istanbul -- but Qatar is resisting the mediator role. The collapse comes as Trump’s April 6 deadline for Iran to reopen the Strait of Hormuz approaches with no diplomatic channel open.

Missing Crew Member: Search Continues Behind Enemy Lines

Six-Word Story: Day two. Still searching. Still hoping.

The weapons systems officer from Friday’s downed F-15E remains missing in Iran’s Kohgiluyeh province. Iran’s IRGC is conducting a ground search and regional officials continue to offer rewards for his capture. The rescued pilot is receiving medical treatment at an undisclosed location.

Justice Alito Secretly Hospitalized Last Month

Six-Word Story: Oldest justice. Hospital visit. Big implications.

Supreme Court Justice Samuel Alito, 76, was treated at a Philadelphia hospital on March 20 after falling ill at a Federalist Society dinner -- an incident the Court did not disclose until this week.^4 The Court says dehydration; sources say he did not faint. But the revelation fuels ongoing speculation about whether Trump could get a sixth Supreme Court seat. Needs context.^5

UConn Heads to the Championship Game -- Again

Six-Word Story: Third title game. Four years. Dynasty.

Dan Hurley’s Huskies beat Illinois 71-62 in the Final Four on Saturday night to reach their third national championship game in four years. Freshman Braylon Mullins -- who hit a 35-foot buzzer-beater to knock off Duke in the Elite Eight -- delivered another dagger 3 with 52 seconds left. Tarris Reed Jr. had 17 and 11. They’ll face Michigan on Monday night for the title, and if they win, Alex Karaban becomes the first three-time champion since UCLA’s dynasty in the 1960s and 70s. Hurley is trying to become the only active coach with more than two championships. “We’re a tough program, we’re a group of fighters,” he said. The 19-game streak in Sweet 16-or-later rounds is the longest active run in college basketball.

NASA’s Livestream Is the Best Show on Television

Six-Word Story: No script. No ads. Just space.

Forget the streaming wars. NASA’s continuous livestream from inside the Orion capsule has become the week’s unlikely cultural event, with tens of millions watching four astronauts float, work, joke, and stare out the window at a Moon that’s getting bigger by the hour. Christina Koch reported seeing the Moon through the docking hatch: “We all had a collective expression of joy.” The crew woke up Saturday to Chappell Roan’s “Pink Pony Club” -- NASA picks the wake-up music. Sometimes the best show isn’t on a screen. It’s out a window.

Japan’s Tanker Braves the Strait

Six-Word Story: First Japanese ship through. Others watching.

Japan’s Mitsui OSK Lines confirmed that the LNG tanker Sohar LNG crossed the Strait of Hormuz on Friday -- the first Japan-linked vessel and the first LNG carrier to transit since the conflict began. The crossing suggests Iran’s selective reopening may be expanding beyond its initial list of friendly nations.

What Jon, Jonah, and the World Might Say

Jon Stewart -- The Liberal’s Honest Reckoning (as imagined by Unspun)

Tomorrow, kids are going to hunt for Easter eggs on the White House lawn, and 20 feet away, there’s going to be a NASA booth where they can walk on a fake Moon while four actual humans orbit the real one. That’s the most accidentally poetic thing this administration has ever done. And I guarantee nobody planned it. The same week we lost an F-15 over Iran and a pilot is missing behind enemy lines, we’re also farther from Earth than any human has ever been. Both things are true. Both things are America. We can reach the Moon and we can’t find the off-ramp to a war we started five weeks ago.

Jonah Goldberg -- The Conservative’s Honest Reckoning (as imagined by Unspun)

The ceasefire collapse is a self-inflicted wound. Pakistan, Turkey, Egypt, and Qatar -- four countries that owe us nothing -- all tried to broker talks. Iran said no, but here’s the conservative problem: the April 6 deadline to “obliterate” Iran’s power grid makes negotiation impossible. You can’t extend an olive branch while publicly sharpening a sword. Reagan talked to the Soviets while building up the military -- but he didn’t tweet a countdown clock to nuclear annihilation. Deadlines that you can’t enforce and can’t walk back are worse than no deadline at all.

The Global View -- What the Rest of the World Sees (as imagined by Unspun)

Japan’s tanker crossing the Strait of Hormuz is the quiet story with the loudest implications. Until Friday, Iran was only letting Russian, Chinese, Indian, and Pakistani ships through -- a curated list of geopolitical allies. Japan isn’t on that list. If Iran is quietly expanding access, it suggests Tehran is calculating that a total blockade hurts its own leverage more than a selective one. Every additional country that transits safely weakens the case for a U.S. military response to “reopen” the strait. Iran may be winning the economic war by slowly making it irrelevant.

Unspinning the Claims, Left and Right

1. Artemis II sphere-of-influence entry confirmed by NASA. All mission data from NASA’s official flight blog

2. 252,757-mile distance record and Apollo 13 comparison confirmed by NASA and ESA.

3. Ceasefire collapse confirmed by WSJ, NDTV, Times of Israel.

4. Alito hospitalization confirmed by AP, WSJ.

5. Needs context. Retirement speculation based on age and this single incident, not any statement from Alito. He has resumed full duties. The Court’s two-week delay in disclosing the incident is fueling the story more than the incident itself.

About the Unspun News

Every morning, we take the spin off the news. We read what the left is saying, what the right is saying, and then check the facts with wire services and the best balanced and international outlets to find what’s actually true. We imagine what Jon Stewart (the left), Jonah Goldberg (the right) and the rest of the world might say. Disputed claims are flagged. You decide what to think. By Mark Palmer and his Unspinning AI Agent.

Subscribe the Unspun News here:

The Unspun News

Every morning, we take the spin off the news. We read what the left is saying, what the right is saying, and then check the facts with wire services and the best balanced and international outlets to find what's actually true. Then unspin it.

By Mark Palmer

90 more?

I stayed with it. I found new angles. I zoomed in. I walked back 100 feet. I played with aperture and shutter speed. Most of the pictures were average. A few stopped me cold.

The exercise wasn’t about photography. It was about seeing.

Seeing with AI

AI is far more powerful than a camera, but the relationship is the same. It’s a tool; you are the creator, the seer, the one who judges its work. You are the soul of the AI machine.

And as a parent, you’re the guide for the next generation of AI creators, seers, and judges. Your kids are already AI-native. They already know how to use the camera. Your job is to help them learn to see.

Forget the robots. One of the best examples of AI in action is the quiet, beautiful work of proteins.

The 50-Year-Old Science Problem

For half a century, scientists have been mapping protein structures. These tiny molecular machines run nearly every process in your body. Understanding their shape is the key to curing diseases, designing new materials, and cleaning up pollution.

After 50 years, the world’s scientists had collectively mapped about 180,000 proteins. A PhD student might spend their entire four-year academic career mapping a single protein. Progress was glacial.

Then came AlphaFold.

In about a year, Google DeepMind’s AI predicted the structure of over 200 million proteins. It didn’t just inch the work forward; it revealed an entirely new world.

And this is where the fear comes in. AI did in one year what took humanity 50 years. Is the scientist’s job over? No. Their job just got a much better camera.

The 200-Million-Book Library

AlphaFold hasn’t replaced scientists. It stood on the shoulders of the 180,000 structures that humans had found. It learned the language of biology from their work. Then, it built a new library with 200 million new volumes. Each describes the predicted shape of a protein that was previously a mystery. But these are predictions, not proof. AI has given us 200 million new images of the world. But it can’t decide which ones matter and which ones don’t.

AI can gather, create, and catalog books, but it can’t read them. It can’t be moved by them. It has no curiosity. It doesn’t know which book holds the cure for Alzheimer’s, or which one describes an enzyme that can eat plastic. And it can’t prove that any of them are real. That takes humans.

What This Means for the Future of Work

That library of 200 million protein structures isn’t sitting on a dusty shelf. Today, over three million researchers in 190 countries use it to accelerate their work.

It’s driven a 40% increase in the submission of novel experimental protein structures. It’s spawned entirely new companies. Isomorphic Labs, for example, uses AlphaFold’s insights to design drugs. The AI camera, in the hands of millions of scientists, is fueling a wave of discovery and the jobs that come with it.

And proteins are just one example. AI is generating new possibilities like this everywhere, right now. AI helps doctors detect breast cancer up to 40% earlier than traditional screening alone. AI helps lawyers research case strategy in minutes instead of weeks. AI helps programmers write better, faster code. AI helps marketers test and refine messaging before spending a dime. AI helps wind turbines adjust their blades in real time to squeeze more energy from every gust. AI helps farmers predict crop disease before they spread. AI helps architects design buildings that use less energy before a single brick is laid. AI helps financial institutions catch fraud in milliseconds. AI helps teachers build personalized lesson plans for every student in the room. AI helps climate scientists model wildfire behavior to protect communities before the flames arrive.

Yes, according to the World Economic Forum, AI is projected to displace 92 million jobs by 2030, like some repetitive lab work that was once needed to map basic protein structures by hand. But at the same time, AI will create 170 million new jobs, a net gain of 78 million.

And there are real risks in every one of these applications. AI can be misunderstood, misused, and weaponized.

And that’s where you come in.

Three Ways to Teach Your Kids to See

Your knowledge, your experience, and your wisdom are exactly what AI is missing. Here are three versions of my “100 pictures of a flower” exercise you and your kids can try today.

Make 100 Pictures of a Flower with AI. Open an AI image generator1 and give it a prompt: “a flower.” Then do it 99 more times: “A flower on the moon, in black and white.” “A single flower in the style of Miro.” “A flower in a Da Vinci sketchbook.” “A massive flower next to a small woman, modern and cool.”

The difference between a simple prompt and a sophisticated one is vast. For example, instead of ‘a flower,’ a more experienced user might write something like this:

A Post-impressionist nature morte in the style of Paul Cézanne, oil on canvas with thick impasto brushstrokes and geometric forms. A lush cacao tree stands on the left with a golden lyre of Apollo leaning against its textured trunk. On the right, a dark wooden table features a blue-and-white Chinese porcelain cup filled with rich liquid chocolate, with an owl perched beside it, peering curiously inside. A spilled bag of white crystallized sugar lies on the table, with the eagle of Jupiter perched upon the sugar. Earthy color palette with patches of vibrant blues and deep browns, structured composition, soft studio lighting, high resolution 

I used the same prompt but made it about a flower, and got this:

Most of the results will be boring. Some will delight you. But the point is learning to see possibilities, to develop your own taste, to find your own point of view. I wrote about this exercise in detail in The Joy of Generative AI Cocreation. It’s the single best way I know to shift someone from fearing AI to playing with it.

Have a Conversation with a Person Who Isn’t There. Ask AI to assume the personality of Abraham Lincoln, Marie Curie, or your favorite author. Let your kids choose who to talk to, from Taylor Swift to Tyler, the Creator.

Then ask them a real question. How would Lincoln handle online bullying? What would Curie think about AlphaFold’s 200-million-book protein library? What would Tyler think about Bad Bunny’s Super Bowl performance?

This isn’t a parlor trick. It’s a way to teach kids how to think about a problem from a different perspective, using AI as a dialogue partner rather than an answer machine.

Create a Marketing Campaign That’s Uniquely Yours. Pick a product, say, a healthy pizza, and ask AI to help you build a campaign around it. Generate images in the style of Salvador Dalí. Write taglines in the voice of your favorite comedian. Ask AI to write some song lyrics in their favorite style.

Then, work to apply their sensibilities to the output. Keep generating more. Use the initial results to create new ideas, or combine ideas like this promotion for “healthy pizza” that combined an AI-inspired slogan, “Because Dali Wouldn’t Eat Boring Pizza,” with a quirky animated pizza image made of peppers that looks like Salvador Dali.

The exercise teaches kids that AI amplifies what makes them unique. The weirder and more personal the prompt, the better the result.

All three of these exercises teach the same lesson. As Mr. Rogers once said, “Nobody else can make the sound that you make.” The same is true with AI. Nobody else can make the prompt that you make and guide AI to novel ideas. That’s the whole point.

The Soul of the Machine

You are the photographer. You and your kids are the explorers who will use those 200 million proteins to find novel cures for diseases. You’re the ones who will use AI to design buildings that breathe, write code that solves real problems, and catch fraud before it hurts real people. You won’t do it by memorizing answers. You’ll do it with the one thing the machine will never have: a burning question. A creative spark. The ability to see.

This is what it means to be the soul of the AI machine. The machine provides the scale. You provide the sight.

The future of work isn’t a race against the machine. It’s a partnership. The most valuable skill you can teach your children isn’t the answer to a test, but the courage to ask a great question. The jobs of the future will go to the curious, the creative, the ones who can walk into this vast new library and know which book to pull off the shelf.

Your kids already know how to hold the camera. Help them learn to see.

Check Out These Substacks I Love

Some of my favorite Substacks on writing, creativity, and its relationship with technology:

• The Intrinsic Perspective by Erik Hoel

• The Story Club by George Saunders

• Polymathic Being by Michael Woudenberg

• The Muse by Birgitte Rasine

I’m editing a new Substack for AI One called Enterprise Context Management. This is a new field of technology that’s all about the context, or instructions and guidance, you provide AI in an enterprise. Follow that Substack and the cofounders Fergus Keenan and Conor Twomey.

And my favorites on AI, enterprise data, and semantic layers:

• Michael Spencer’s AI Supremacy

• Marcus on AI by Gary Marcus

• Last Week in AI

• The Algorithmic Bridge by Alberto Romero

• The Pragmatic Engineer by Gergely Orosz

His name was William Zinsser, and he taught a generation how to write with clarity and strength. Though he died in 2015, I’ve resurrected his ideas with artificial intelligence. I feed AI my drafts, and Zinsser’s voice comes back to me, sharp and clear, telling me to cut, to simplify, to make every word work.

This is the secret of writing with AI. It’s not about generating text; it’s about thinking. It’s not a replacement for the writer, but a powerful new tool in their kit.

Editing with Zinsser is just one way I use AI. My methods surprise even experienced writers. I use AI to explore ideas, to find my voice, to scramble words like David Bowie’s Verbasizer. To do this work, you need the right mindset. You must think of AI not as a machine that writes for you, but as a rock tumbler that polishes your thought.

The Great “AI is Faster” Lie

How much time do you spend on an important piece of writing? In a workshop I teach, 80% of students said they spend less than five hours.

That’s not enough.

Nancy Duarte, an authority on presentations, says a one-hour talk should take 36 to 90 hours to create. Most of that time is spent thinking and researching, not writing slides.

Writing is getting worse thanks to AI; those same students estimated they spend 50% less effort when they use AI.

The idea that “writing with AI is faster” is a dangerous illusion. The evolved craft of AI writing uses it as a tool to tumble our ideas, turning jagged thoughts into polished prose. It helps you expend more effort, not less, by opening up new realms of thought.

Explore Possibility Spaces

The philosopher Tobias Rees describes AI as a tool to create “Possibility spaces that humans can travel but not produce on our own.” Instead of just finding information, AI-savvy writers prompt the machine to explore and connect ideas, revealing new landscapes of thought.

In RIP, BI. The Death and Rebirth of Thinking, I shared an example of using AI to explore a product idea. On my way to the airport, I wrote a 700-word prompt. In the security line, I sent it to an AI research tool, Manus. Here’s the prompt:

As I waited, Manus read over 20 books and patents. It analyzed my idea. It found prior art. It generated counterintuitive concepts. It made a podcast that debated my idea. AI didn’t produce finished writing, it created raw material. A quarry of ideas, waiting to be shaped.

Find Your Unique Voice

Good writing is good thinking. It’s personal. It’s passionate. AI can help sift, sort, and grind your ideas until they reveal your unique sound. The popular view is that AI produces generic slop. The opposite is true. Used correctly, AI makes them more personal, not less.

I like to play with ideas by expressing them in different forms. I ask AI to generate images in the style of my favorite artists. I ask it to write Hemingway six-word stories. Anita Stubenrauch uses it to write haikus.

This kind of prompting helps you find what Wes Kao calls a Spiky Point of View. It’s your take on the world that’s surprising, true to you, and sharpened by the vast possibilities AI reveals.

Which spiky idea to focus on is entirely up to you; AI helps you find and develop them.

How to Edit with the Dead

Here’s how I write with Zinsser. I paste a draft into an AI and give it a simple prompt:

Evaluate this draft from the perspective of William Zinsser, the author of On Writing Well, and give me suggestions on how to make it better.

The AI assumes Zinsser’s voice. It gives me his core principles: eliminate clutter, use strong verbs, be direct:

AI's ability to assume a personality is one of its most powerful features. It helps you see your thinking from another point of view. I favor Zinsser's voice, but you can substitute any writer whose style you admire, and the AI will channel their perspective on your work.

Verbasize: Prompt Yourself

Most view AI as something you prompt; AI-savvy writers use it to prompt themselves. It’s like a tool David Bowie used over thirty years ago called the Verbasizer to randomly reorder words, sparking new lyrical connections. In Michael Apted’s Inspirations documentary, Bowie described how the Verbasizer would take his raw ideas, cut them up, and randomly recombine the words to create new connections:

Verbasizing is like controlled hallucination. AI's tendency to make unexpected connections, to leap from one idea to another in ways a human might not, becomes a creative tool. Instead of a bug, it's a feature. You're not looking for facts; you're looking for sparks.

Bowie would have loved AI. It’s an exponentially more powerful version of his invention. Feed AI any raw material (a sentence, a concept, your spiky point of view) and ask it to verbasize new combinations. New perspectives. New interconnections between concepts. I used AI to verbasize my airport product exploration like this:

Read 100 patents, journals, blogs, and technical posts that have similar ideas to mine. Compare them to each other and find 20 novel insights and connections.

Bowie said the Verbasizer was like a “technology dream” that gave him access to new ideas. Using AI to prompt yourself is one of its superpowers. It’s another turn of the rock tumbler, revealing surprising new facets of your thinking.

The AI Writer’s Toolkit

The AI-savvy writer knows the machine is just one tool in the box. The craft still depends on the human touch: the hard work of choosing every word, the discipline of reading a sentence aloud to hear its rhythm, the courage to seek feedback from other people.

My own process moves between the digital and the analog. I often think on paper, where research shows we generate more ideas. Here’s a journal entry where I first sketched the rock tumbler concept.

Then I move to the computer to draft. Each tool has its purpose. The goal is not to replace one with another, but to build a process where they work together.

Research supports this approach. Studies show that when humans use a pen and paper, we generate more ideas than typing on a computer. AI multiplies our ideation power. But computers are better for human collaboration.

So the complete workflow is: paper to think, AI to tumble your ideas, and computers to collaborate with others.

Murder Your Darlings? AI Swings an Ax

“Delete this entire section. You already told this story. Cut all of it. Be ruthless.” 

This was advice from my AI editor after I rewrote the introduction.

The phrase “Murder your darlings” traces back to the British writer Sir Arthur Quiller-Couch in 1916. It’s a command to edit without sentiment. If a sentence, however beautiful, does not serve the whole, it must be cut. The act feels like murder, which is why it is so hard and so necessary.

I loved the section on editing with Zinsser. It was my darling. AI told me to kill it. A human editor might have softened the blow. AI did not. It was direct. It was brutal. And it was right. The section was redundant.

But in the stark clarity of AI’s command, I got an idea. The section had a valuable lesson sandwiched inside the redundant story. So I didn’t kill my darling, I operated on it. I transformed the section into a practical guide: “How to Edit with the Dead.”

AI’s brutal feedback didn’t force me to cut. It forced me to question.

Build Your Own Master Class

The final step in my writing process is to grade my own work. Not on a whim, but against the principles of the masters I admire. I use my AI editor to build a personal master class, with a faculty of my own choosing.

First, I asked it to synthesize Zinsser’s principles into a 20-point writing rubric. This became my final exam for the article you are reading. The AI graded the piece on principles like “Simplicity,” “Humanity,” and “Strong Lead.” It’s a test of the article’s soul.

But philosophy is only half the writing battle. For the other half, I turned to Roy Peter Clark, the author of Writing Tools: 55 Essential Strategies for Every Writer. Where Zinsser is the philosopher, Clark is the craftsman. I had AI create a second rubric based on Clark’s tools, covering the practical nuts and bolts of the craft. It graded my article on techniques like “Emphatic Word Order,” “Ladder of Abstraction,” and “Gold Coins.”

This is how you use AI to create a personalized master class. You are no longer limited to the feedback of one editor or one teacher, and it’s free. You can assemble a faculty of the greatest writing instructors in history, synthesize their wisdom into objective rubrics, and get a comprehensive assessment of your work. You can ask Zinsser if your writing has heart, and then ask Clark if your sentences have muscle.

I’ve included the Zinsser and Clark rubrics, along with the final scoring for this article, as a downloadable PDF so you can build your own master class:

Scoring The Dead Man Against The Zinsser Rubric

75KB ∙ PDF file

Download

Download

Scoring The Dead Man Against The Roy Peter Clark Rubric

123KB ∙ PDF file

Download

Download

Scoring The Dead Man Article Against The Zinsser Rubric

75KB ∙ PDF file

Download

Download

The Companion in the Machine

A friend recently told me about his father, who has dementia. The father is a World War II veteran, and all he wants to do is talk about the war. He tells the same stories, ask the same questions, for hours on end. For the family, it’s exhausting.

They set up an AI companion for dad. It listens patiently. It asks thoughtful questions. It can discuss the Battle of the Bulge all day. The father is engaged and content. AI lifted the burden of being a 24/7 conversational partner. AI doesn’t replace the family’s love; it does crucial, compassionate work that was previously going undone.

This example reframes one of the most contentious debates about AI: Is it out to replace human work? Most think yes; I see a spectrum that’s mostly good news for us humans. Yes, it will take a few jobs. But mostly, AI will create new opportunities, new tools, and even new companions.

This article went through 36 revision cycles, about 1,600 discrete editorial suggestions, and was scored against 70 principles from two master writing instructors. In a traditional editorial process, this level of collaboration would have taken 9 to 18 months of elapsed time and cost $8,500 to $12,750 in professional fees. I did it in a week.

The bottleneck in traditional editing has never been the quality of feedback, it's the waiting. AI eliminates the wait. Every suggestion arrives in seconds, not days. Every revision cycle completes in minutes, not weeks. This isn't about replacing human editors. It's about making a level of editorial rigor accessible that was previously reserved for bestselling authors with publishing contracts and unlimited budgets.

I never would have spent that money. I’m writing this for the love of it. AI gave me an opportunity to push my craft to a level. It allowed me to do the work that, for a personal project, would have gone undone.

A Call to Tumble Rocks

That’s the AI writing rock tumbler. You gather jagged stones of your ideas. You explore new landscapes AI reveals. You verbacize. You select the most promising ideas and tumble them with the help of great masters. The hard work of writing is the grit that brings out your inner character.

What emerges is not a machine’s product, but a gem of human thought, hardened and clarified by AI, other people, and the creative sparks that AI helped you find. This is the opposite of using AI as a writing machine.

Want to try it? Here’s how:

1. Change Your View of AI. Stop using AI to generate words; use it as a rock tumbler instead. Learn to use it as a possibility space explorer, an editor, a verbacizer, and a critic. Use it to fuel your own unique thoughts, not to appropriate what other people have already said and shared.

2. Write Slower. Reject the AI writing speed myth. AI writing is better, not faster. Explore new frontiers. Wonder more. Edit slowly. The goal is not to be first. The goal is to be remembered.

3. Practice New Ways of Using AI. Learning to use AI is like learning any new, complex tool. It requires patience and practice. This article is a map, but you must walk the territory yourself. Each AI is different. Each writer has a spiky point of view. Learn from the writers you admire. Learn the nuances of the machine, its strengths and its weaknesses, before you can bend it to your will.

Survival of the Fittest Writer

Writers who thrive in the AI age will use it to enhance their craft, not replace it. They will understand that good writing comes from good thinking, and that good thinking cannot be automated. They will use the rock tumbler to polish their ideas, patiently and methodically, until they shine.

The writers who fail will be those who take shortcuts, becoming mere operators of a writing machine, spewing out content that is instantly forgettable. In an attention economy, that is a death sentence.

The choice is yours. It is the choice every species faces when its environment changes. Adapt or perish. You can produce content that wastes everyone’s time, or you can create something worth reading. Your readers will know the difference. More importantly, so will you.

Gratitude

Each year, I donate all Substack subscription fees to Care Dimensions, the hospice that cared for my family when my wife died of cancer. Your support has raised thousands of dollars for them. Thank you.

Subscribe now

Check Out These Substacks I Love

Some of my favorite Substacks on writing, creativity, and its relationship with technology:

• The Intrinsic Perspective by Erik Hoel

• by

• Polymathic Being by Michael Woudenberg

• The Muse by Birgitte Rasine

I’m editing a new Substack for called Enterprise Context Management. This is a new field of technology that’s all about the context, or instuctions and guidence, you provide AI in an enterprise. Follow that Substack and the cofounders and .

And my favorites on AI, enterprise data, and semantic layers:

• Michael Spencer’s AI Supremacy

• Marcus on AI by Gary Marcus

• Last Week in AI

• The Algorithmic Bridge by Alberto Romero

• The Pragmatic Engineer by Gergely Orosz

The premise is simple. An enterprise thinks by drawing on its collective memory of successes, failures, and principles. This is its context. For AI to be effective, it must tap into this “enterprise brain.”

The breakthrough is this: AI needs tools to apply the right context to the right prompt at the right time. ECM helps companies tap into that enterprise brain. This breakthrough ushers in the era of the thinking enterprise, powered by AI reasoning at scale. Read the introduction to ECM on the Substack here:

Enterprise Context Management

Your AI is Blind: Enterprise Context Management Arrives

Your company has spent millions on infrastructure with the promise of AI. Business impact still eludes you. What’s missing isn’t more infrastructure, it’s the ability to activate your data so AI can drive real outcomes…

Read more

8 months ago · 28 likes · 9 comments · Fergus Keenan and Conor Twomey

While prompt and context engineering are on the right track, they are incomplete. They lack a systematic way to organize context, an idea that Mark Sykes explains in his “Context Skyscraper” essay.

Enterprise Context Management

The AI Context Skyscraper

For AI, great answers come from more than great questions. They come from great context. The right context includes the guardrails, historical data, environmental information, and guidance that turn raw outputs into real insights…

Read more

7 months ago · 16 likes · 4 comments · Mark Sykes

They lack a method to guide AI models with your principles. This is the “Hybrid Loop.” Read about it here:

Enterprise Context Management

Stop Building Agent Chains. Start Building Hybrid Loops.

If you work with modern AI models, you will soon feel a strange tension. On one hand, you are told to give the model everything in one large prompt. This includes long context, rich instructions, a list of tools, and then you let the model figure it out. On the other hand, the problems you care about are messy. They are long running and tied to systems …

Read more

6 months ago · 15 likes · 1 comment · AI One and Mark Sykes

With $11M in funding, our journey has begun. Follow AI One on LinkedIn, cofounders Conor Twomey and Fergus Keenan and our Substack for perspective on ECM, how it works, and why it matters.

The human role has shifted from craftsman to pilot. Where BI experts once hand-crafted dashboards with Tableau or PowerBI, they now fly a powerful new exploration vehicle. AI agents provide a vehicle for multi-dimensional, sub-atomic data discovery, but the human charts the course. They know which questions to ask, which paths to explore, and which insights matter.

The report’s findings make this shift concrete. It found that AI-native software providers are three times more likely to provide more than 26 external data sources compared to traditional data exploration tools (46% vs. 15%).

This leap in data availability requires human pilots who know the terrain and can navigate with the powerful new tools at their disposal.

Download the full report here: https://lnkd.in/eD53cZcN

Where Have You Been, Mark?

The Techno Sapien Substack has been quiet: sorry! This is one of the projects I’ve been working on — a survey (and related set of deep research interviews) of over 200 leaders in enterprise and SaaS software about what they’re ACTUALLY doing with AI. I’ll be sharing more key findings over coming months.

Also, I helped AI One launch a new Substack about an emerging technology field they call Enterprise Context Management. Subscribe to it here!

Enterprise Context Management

Enterprise Context Management is an emerging technology that helps enterprises extract and apply context, turning everyone’s AI into your AI.

By AI One

ECM is a new field that helps enterprises control the context they give AI to get better results. I’m editing the awesome artless over there by cofounders Conor Twomey and Fergus Keenan, and AI Architect Mark Sykes.

And Check out these other Substacks I love:

• Michael Spencer’s AI Supremacy

• Marcus on AI by Gary Marcus

• Last Week in AI

• Polymathic Being by Michael Woudenberg

• The Algorithmic Bridge by Alberto Romero

• The Muse by Birgitte Rasine

• The Intrinsic Perspective by Erik Hoel

• The Pragmatic Engineer by Gergely Orosz

• Hugh Kennedy from PJA

Today, companies face the same choice with artificial intelligence. Many bolt AI onto existing systems; they’re failing. Others rebuild their foundations to capitalize on AI’s unique capabilities.

The big question is “how?” In my latest eBook for CData Software, I suggest six moves to rewire for AI. They include:

1️⃣ Recompose Data Pipelines for Lightweight Workflows – Move from heavy, batch-oriented pipelines to direct, real-time integrations that feed AI with live customer context.

2️⃣ Embrace Pop-Up Data Integration – Use temporary, on-demand, standards-based connections (based on MCP) that appear when needed and vanish afterward, cutting infrastructure costs.

3️⃣ Move from Data Lakes to Live Data – Shift from static data warehouses to direct, live access that enables real-time AI applications.

4️⃣ Design for Multiple Connectivity Points – Treat broad connectivity as a core feature, building a rich library of connectors to meet exponential AI-driven demands.

5️⃣ Design Data for Codegen, Not Manual Work – Structure your data to be machine-readable and compatible with AI code-generation platforms, accelerating development.

6️⃣ Shift Application Targets from BI to AI – Reorient data services to serve AI agents directly, moving beyond traditional dashboards to embed conversational insights.

Download the full eBook from CData that explores these six moves: “Six Moves to Rewire Software for the AI Age.”

Where Have You Been, Mark?

The Techno Sapien Substack has been quiet: sorry! I’ve been busy writing for other people, as in this case with the CData eBook.

Also, I helped launch a new Substack about an emerging technology field they call Enterprise Context Management. It’s about how to control the context you give AI to get better results. This is yet another form of rewiring the enterprise! I’m editing the awesome artless over there by cofounders and , and AI Architect . Subscribe to it here: .

Check out these other Substacks I love:

• ’s AI Supremacy

• Marcus on AI by

• Last Week in AI

• Polymathic Being by Michael Woudenberg

• The Algorithmic Bridge by Alberto Romero

• The Muse by Birgitte Rasine

• The Intrinsic Perspective by Erik Hoel

• The Pragmatic Engineer by Gergely Orosz

• and Hugh Kennedy from PJA

Connect on LinkedIn

See you there!

"How many of you LOVE AI?" I asked.

Ten percent raised their hands. Tentatively.

"How many loathe AI?"

Over half the hands shot up. Vigorously.

I knew my defense of AI was doomed.

The Talk That Started a War

My talk (here’s its essence) raised hackles and hell. Despite my attempt to share a positive message, the loathers were unconvinced. During a post-talk roundtable, the moderator asked how many of the AI loathers were swayed by my talk. Not a single hand.

Online, attacks turned personal. My talk was called “The Worst I’ve Ever Seen.” People I have never met blocked me. I felt like a Trump supporter hoodwinked into presenting at a Bernie Sanders rally. The controversy grew so fierce that four industry leaders asked me to join their show, Chart Chat, to address the blowback. Scarred yet hopeful, I agreed.

Here’s what the war taught me.

The Shock: Smart People Hate Smart Machines

When I first saw the Pew Research numbers, I didn't believe that 52% of the population loathes AI. When I presented at Outlier, I expected openness, debate, and support.

My instinct was wrong.

Massively, embarrassingly wrong.

I thought education bred acceptance, but it also breeds contempt. The loudest critics were luminaries. Authors. Academics. Community leaders with a lot to lose.

Maybe success breeds fear and loathing. Maybe fear explains why only six percent of organizations are successfully using AI. Maybe 94% of leadership silently hates AI because it threatens their position of power.

That's tragic.

This insight carries weight beyond my hurt feelings. Investors may be well served to reflect on whether their CEOs genuinely embrace AI or harbor reservations about it. CEOs might need new blood to lead AI initiatives. Everyone should wonder if their objections to AI are real or based on fear. AI loathing can be economically devastating, competitively suicidal, and career-limiting.

Lot of Spite, No Action

My mom used to have a sign on our refrigerator that read, “Complain about anything as long as you have two solutions, too.” Her wisdom echoed in my mind as I responded to comments after my talk.

I fielded many complaints about AI. Its unethical consumption of power. Disgust with Mid Journey-generated images. Annoyance with AI-generated slop on LinkedIn. All fair criticisms. But the complaints weren’t coupled with solutions. Their hatred, hollow. The strongest objectors had the weakest ideas about AI and its impact on human work.

Each time, I offered constructive solutions. One asked accusingly, "Aren't you concerned about AI bias?"

"Yes! I coauthored a patent for identifying and mitigating AI bias, involving human adjudicators to address it." Nobody asked a follow up. Nobody was interested in solutions.

Crickets.

Talk of solutions seemed to reveal that the loathers weren’t looking for answers; they were just out to play the problem-pointing game.

The AI Creativity Wars

Is prompt-writing creative? I asked. This question was the center of my talk, about how AI can amplify human creativity.

More crickets. But I noticed several nodding heads too. I felt the quiet support.

Upon reflection, this question struck the heart of the identity of the most offended folks in the room. Maybe my question questioned their worth. Ironically, this is the opposite of what I was saying.

The most frustrating debates centered on abstract ideas: AI intelligence (it's not); AI enhances creativity (it does); using AI is a craft (it is).

I presented a 599-word prompt I crafted for in-depth patent research. The prompt took weeks to write, leans on thirty years of experience in product invention, company leadership, and go-to-market strategy. The idea that creative prompting is an art and science isn’t original. Everyone that uses AI aggressively knows that the more you put into your questions, the better the output. This post, “The $20,000 Growth Consultant Prompt,” went semi-viral on LinkedIn. Its author claimed the wisdom in this prompt was worth $20,000. I didn’t go that far with mine!

An AI Loather objected loudly. And proved my point accidentally. "But when I tell AI to 'Improve my writing,’ I get useless garbage!"

Exactly. Thanks for making my point.

An Offensive Example

One leader picked apart this Midjourney-generated trumpet. He wrote in all caps that he HATED the image. He HATED the fake plastic look of the valves. He HATED its style. But the trumpet image distracted him from the point I was trying to make: that AI can help humans be more empathetic.

Empathy and connection stem from knowing others — not just their demographics or appearance, but also their daily frustrations, their language, and their fears.

“Persona assumption” is using AI to provide answers from a different point of view. At home, you make AI sound like Yoda or Seinfeld. At work, make it think like your best customer. The image I shared is a simple manifestation of persona assumption: to imagine a trumpet as artist Jean-Michel Basquiat might imagine it. Salespeople can use AI to interpret their pitch from the prospects’ point of view. Lawyers can use AI to understand a jury’s interpretation of their argument. Marketers can understand how their audience might find their content manipulative.

My message was about using AI as an essential instrument to find empathy; I wound up fighting about plastic-looking trumpet valves. Maybe the derailment caused by my AI-generated image wasn’t worth it. On the other hand, some people loved it. Next time, I’ll keep the trumpet.

The Prior Art Paradox

“If you use AI, you’re stealing.” Another war erupted. This one raised my hackles, because it isn't just criticism, it's a moral indictment of AI users.

Questions about creative rights and intellectual property started the discussion. These are fair and valid concerns, and I acknowledged the issue. I returned to the central point: using AI is hard work, and users should never use AI’s output unmodified. I also explained how hard AI toolmakers work to protect IP rights. I brought up the lessons learned from the New York Times dispute, especially OpenAI’s fascinating response, which outlines the many concrete ways they work to stamp out theft and empower publications to control how their content is used, or not, during training.

But the loathers kept returning to the same objection: AI steals.

I enlisted a quote from Rick Rubin for support. Rubin produced Grammy Award-winning music by Adele, Johnny Cash, the Red Hot Chili Peppers, and Metallica. He wrote,

"Nothing in the world is entirely new. All art is, in some sense, a remix or a continuation of what came before. We are of the world. We take in the world, combine it with the self, and express it in a new way." 

Originality is overvalued. Being open to inspiration from any and all sources is the real work. The work is to make combinations no one else has thought to make—to see connections others don't—until something new comes into being that only you could have made." 

Rubin uses AI for inspiration. He uses it to make connections. He sees it as a tool to express original ideas built upon what came before. This argument, too, went nowhere. Even Rick Rubin couldn't sway the haters. If a legend can't convince them, who can?

This was my final moment of despair.

My Opening Farewell to the Loathers

The haters were loud. The others were many. I was spending more physical and emotional time with the vocal, negative minority instead of the silent, productive majority. The minority blocked the possibility of progress with the majority.

For decades, I’ve listened to Jackson Browne when I’m sad. One night, while responding to a LinkedIn Loather with a glass of wine, his song "My Opening Farewell" counseled me:

But suddenly, it’s so clear to me,
That I'd asked her to see what she may never see.

Browne was singing about a girl; I was writing to loathers. Suddenly, I realized that trying to convince AI loathers is wasted time. I'm asking them to see what they may never see.

Thanks, Jackson Browne, for encouraging this opening farewell to the AI loathers.

Moving On With AI

I’m moving on. I wasted time fighting the AI Loathers, but I also made new friends and colleagues. At Outlier, I shared three steps to embrace AI. I still think they’re sound.

1. Treat AI prompting as a craft. The more time, energy, wisdom, and care you invest in AI prompts, the better results you get.

2. Embrace AI hallucinations. For creative work, hallucinations are unfairly maligned. If you aggressively consider, question, challenge, and embrace AI output, it can spark new insights and elevate human work. AI can also be wrong. Devour those synthesized morsels. Chew them up.

3. Work hard when you work with AI. Using AI is hard work. Good work. Honest work. Critical thinking work. Consuming, questioning, selecting, editing, combining, and exploring its suggestions is a creative act. But own the responsibility for any bias, errors, or poor choices when you use them.

Using AI in this way, like many things, takes a minute to learn and a lifetime to master.

I learned something from Jackson Browne: Sometimes the best way to win a war is to stop fighting it. Loathers will loathe. Hypsters will hype. In between, in that vast middle ground, that's where real change happens. There’s a lot of work to do, and I’m moving on. One prompt at a time.

AI, when used with a semantic layer that encodes corporate knowledge, can evolve the craft of analysis from query-answering to insight discovery, fundamentally changing the user's interaction with data.

Analysts Are Evolving

Analysts are evolving from query facilitators to insight discoverers.

For example, instead of a marketing manager using a business intelligence tool to ask questions about campaign performance, they might first use AI with a semantic layer to unearth connections that had previously gone unseen, like "Campaign X seems to be underperforming in the Midwest region due to ad creative misalignment with local preferences, as suggested by weak campaign KPI performance and higher than average negative sentiment from agent conversations."

The semantic layer, according to Mariani, provides the structured context (campaign definitions, demographics, metrics) that AI can use to connect dots and spot anomalies. Without semantic context, AI may surface interesting but business-irrelevant findings.

The Extended Search Party

As Andy Clark and David Chalmers suggest in their essay, The Extended Mind, "Your mind includes your tools." AI is a tool that extends our mind beyond individual cognition. When designed well, semantic layers and AI systems can be part of our extended cognitive apparatus for navigating information haystacks.

Semantic layers help AI guide us toward insights buried deep in haystacks of research, data, and language. Deciding which insights matter is up to you.

This post is a part of a multi-part series about ideas shared in the 2025 Semantic Data Layer Summit. It explores topics like: Is data visualization dying, or is it just getting started? How can AI best leverage data to ensure its insights are accurate? What role do semantic layers play in the agentic applications of today and tomorrow?

To grasp the impact of this claim, consider a marketing analyst who needs to understand recent customer churn trends in her company.

Instead of asking a data engineer to hand-craft complicated SQL queries to join multiple tables, she now crafts a rich prompt for AI. She can ask, in plain English: "Show me the top 5 reasons enterprise customers churned last quarter, correlated with support ticket volume. Check trends for each reason over the last three months, and draft a report summarizing your findings."

The AI reasoning model interprets her question. It asks the semantic layer to map business terms like "Customers" or "Ticket Volume" to underlying physical database structures. Data is retrieved and integrated into the AI workflow. The user gets an answer. Effectively, this new workflow eliminates the need for humans to write SQL. The new paradigm seems clear: "Ask in English, not SQL."

The SQL Scribe: AI or Human?

But here's the rub: SQL isn't dead. Ironically, it's more critical than ever in the AI era. Semantic layers, vibe coding, and AI agents all depend on high-quality, high-fidelity SQL. Therefore, the question isn't if SQL is needed, but who writes it.

AI models don't magically understand your data. They need a precise and well-defined language to interact with it. That language is SQL; semantic layers act as a sophisticated translator, steering the creation of SQL queries needed to pull the right data.

This isn't the end of SQL; it's the end of manual SQL writing for most users. Data professionals, however, now face a new challenge. They must ensure the underlying SQL generated by AI is robust, secure, and scalable. They must also build the high-fidelity semantic models that guide the AI's SQL generation.

So, while analysts can now ask questions in English, a new breed of data architect or AI engineer ensures that the semantic layers and generated SQL behind the scenes are impeccable. They're the unsung heroes, ensuring the AI's answers are accurate, not just plausible.

The SQL scribe has changed, but the script remains essential.

This post is a part of a multi-part series about ideas shared in the 2025 Semantic Data Layer Summit. It explores topics like: Is data visualization dying, or is it just getting started? How can AI best leverage data to ensure its insights are accurate? What role do semantic layers play in the agentic applications of today and tomorrow?

This isn't just a technical decision—it's a philosophical showdown that could reshape enterprise architecture for years to come. Let’s explore two sides of this debate.

POINT: Semantic Layers Should Be in Your Database (or BI Tool)

Josh Klahr from Snowflake argues that semantic definitions should be stored as first-class database objects, like database tables and views, rather than in external files and configurations. At the 2025 Semantic Layer Summit, he explained how Snowflake’s approach “improves consistency, security, shareability, and integration.”

Tableau Semantics does the same thing. Power BI does the same thing. Google Looker does the same thing.

Conventional Wisdom Challenged: Embedding semantics in a tool challenges the notion that semantic meaning should unify all tools, databases, large language models (LLMs), and applications within an enterprise.

Six-Word Story: "Embed semantics in Snowflake, not elsewhere."

Example: Snowflake semantic views help developers create database objects like a "Revenue" metric that can be accessed by any authorized application (Tableau, Power BI, or custom applications) with the same definition, permissions, and lineage information, eliminating the need for each BI tool to maintain its own semantic layer with potentially conflicting definitions.

Making semantics database-native aims to centralize data meaning close to the data source, leveraging the database's security and object management capabilities for better governance and consistency.

Further Exploration: Check out more from Snowflake's native semantic views.

COUNTERPOINT: Semantic Layers Should NOT Be in Your Database

Embedding semantics in databases and BI tools proliferates multiple, contradictory semantic layers across an organization, exacerbating the very problem they’re meant to solve: inconsistent definitions and multiple versions of the truth. According to Chris Lynch, CEO of Atscale, “Embedding semantics in the data platform is a fundamentally flawed idea and is “almost as bad as coupling it to a BI tool, as Power BI does.”

"A semantic layer must be universal or don’t bother!”

Conventional Wisdom Challenged: Contradicts the trend of database and BI vendors embedding semantic capabilities in their platforms, arguing for independence rather than vendor lock-in and “detached islands of data meaning.”

(Seven) Word Story: "Universal semantic layers build bridges, not walls."

Example: A multinational corporation utilized Snowflake for its financial data, Databricks for its marketing analytics, Microsoft Fabric for its operational data, Tableau, Power BI, and Spotfire. Each platform has a native semantic layer with different definitions of "customer" and "revenue." When executives compared reports across departments, they got conflicting insights and spent weeks reconciling the differences. Ultimately, they realized they needed a platform-independent semantic layer to establish a single source of truth.

Semantic Layer Solution: An independent, universal semantic layer that sits above all data platforms and BI tools provides consistent definitions and metrics regardless of where data is stored or how it's consumed, including for AI. This ensures that all users, applications, and AI agents work with the same business logic, regardless of the underlying technology stack.

The semantic layer debate ultimately reflects a deeper tension in data architecture: the trade-off between platform optimization and universal flexibility.

Snowflake's database-native approach offers compelling advantages in governance, security, and performance, making it an attractive choice for organizations heavily invested in a single data platform.

Universal semantic layer advocates argue that independence is the whole point of sharing semantics. The idea is to scale meaning, not trap it in one place, creating vendor lock-in and the proliferation of competing definitions across heterogeneous environments.

What's certain is that doing nothing—allowing semantic chaos to persist—is no longer an option. The key is making a deliberate choice and committing to the discipline required to make it work. Ultimately, the best semantic layer is the one you’ll use.

Further Exploration: Check out Chris Lynch on embedded Semantic Layers

21 Big Ideas from the 2025 Semantic Data Summit

This post is a part of a multi-part series on big ideas from the 2025 Semantic Data Layer Summit. It explores hot takes on topics like: Is data visualization dying, or is it just getting started? How can AI best leverage “real” data to ensure its insights are accurate? What role do data semantics play in the agentic automation applications of today and tomorrow?

Research shows that semantic layers make AI responses 3X more accurate (54% vs 16%) compared to direct SQL database queries. Unlike BI tools that translate data for human consumption, semantic layers create a universal language that allows AI agents to comprehend and act on business context, relationships, and meaning across disparate data sources.

Six-Word Story

Semantic layers teach AI your lingo.

Conventional Wisdom: Semantic Layers are Just for BI

BI teams built it. AI stole it. A prevailing view is that semantic layers are primarily designed to make dashboards better. Instead, semantic layers are becoming the essential neural system for AI agents, enabling them to navigate complex enterprise data landscapes without human intervention. A semantic layer provides a bridge between the language of business and the language of data for both AI and humans.

Case Study: Home Depot

At the Semantic Data Summit, Rick Ramaker described how Home Depot utilizes conversational analytics and natural language interfaces to access its semantic data layer, which was initially built for BI. They experimented with alternative approaches, including direct natural language-to-SQL solutions. Rick’s team discovered that "going from natural language to an API call to our semantic layer was by far the best approach," yielding "really, really good results."

Home Depot's semantic layer contains curated data for the entire organization, making it the ideal foundation for AI agents and machine learning systems. Rather than having AI agents try to understand raw data or navigate complex database structures directly, the agents interact with the semantic layer through API calls.

This approach allows Home Depot to:

1. Leverage their existing investment in semantic modeling (their 20 terabyte semantic layer. More on this in a future post!),

2. Ensure AI responses are based on the same "golden set of data" that human analysts use with their BI tools,

3. Maintain consistency between BI and AI-generated insights,

4. Benefit from performance optimizations built into their semantic layer, and

5. Scale AI and BI across all business functions.

Ramaker says that this semantic layer approach to AI has become "a foundational piece of how we do data analytics at Home Depot: natural language talks to our semantic layer." The company is broadening deployment of AI agents that leverage their semantic layer as the knowledge foundation.

Leverage Semantic Layers for AI, Not Just BI

Modern semantic layers designed for AI agents extend beyond traditional business intelligence (BI) metadata by creating knowledge graphs that encode business rules, relationships, and context in machine-readable formats. These systems enable AI agents to understand not only what data means, but also how different concepts relate to each other and what actions are appropriate in various business contexts. This creates a foundation for autonomous AI operations that can scale beyond human-directed analytics.

So, start attaching your semantic layer to enterprise AI to improve the accuracy, trust, and creativity of AI!

21 Big Ideas from the 2025 Semantic Data Summit

This post is a part of a multi-part series on big ideas from the 2025 Semantic Data Layer Summit. It explores hot takes on topics like: Is data visualization dying, or is it just getting started? How can AI best leverage “real” data to ensure its insights are accurate? What role do data semantics play in the agentic automation applications of today and tomorrow?

Further Reading

For more on this idea and semantic layers in general, explore these resources:

21 Surprising Ideas About Semantic Layers, AI Agents, Generative AI, Business Intelligence, and Data Integration from the 2025 Semantic Layer Summit. This the is the master “rolling thunder” post with all the big ideas from the summit

The Techno Sapien Podcast: Semantic Layers 101. If you’re new to semantic layers.

A Benchmark to Understand the Role of Knowledge Graphs on Large Language Model Accuracy: https://arxiv.org/pdf/2311.07509

How Semantic Layers Make GenAI 3X More Accurate: https://www.atscale.com/blog/semantic-layers-make-genai-more-accurate/

The Great Semantic Layer Debate: Five Issues for AI Innovators: https://technosapien.substack.com/p/the-great-semantic-layer-debate-five

Top Semantic Layer Use Cases and Applications (with Real World Case Studies): https://enterprise-knowledge.com/top-semantic-layer-use-cases-and-applications-with-realworld-case-studies/

The Semantic Layer Summit 2025 (Replays): https://www.atscale.com/resource/2025-semantic-layer-summit-on-demand-sessions/

Last week, a panel of prominent leaders explored these ideas at the 2025 Semantic Data Layer Summit. I attended the proceedings and was blown away by the insights and case studies shared by leaders from Snowflake, Databricks, IBM, Thoughtspot, Home Depot, and the host, Atscale, and more.

There were so many cool ideas that I decided to share them one by one, day by day. This post is a “directory” of them — a rolling thunder of ideas from the summit. Each day, I’ll post a new idea and add it to this list (and share on LinkedIn, too).

Each post features a big idea from the summit, the conventional wisdom it debunks, a case study of the idea in action, and an explanation of how semantic layers help (if you’re new to semantic layers, read this or listen to episode #4 of the Techno Sapien podcast, Semantic Data 101).

Here we go!

On Semantic Layers, AI Agents, Generative AI, BI, and Data Integration from the 2025 Semantic Layer Summit

💡Idea #1: Semantic Layers are BI’s Gift to AI

💡Idea #2: There’s a Semantic Layer standardization battle afoot

💡Idea #3: Will Natural Language take us to the promised land of self-service BI?

💡Idea #4: Can AI balance automation and human oversight

💡 Idea #5: Semantics Should Be In Your Database

💡 Idea #6: Semantics Should NOT Be In Your Database

💡 Idea #7: SQL is Dead. Long Live SQL

💡 Idea #8: What is the future of human analysts in the AI age?

💡 Idea #9: AI is a Needle-Finder, Not an Answer-Giver

💡 Idea #10: How can semantic layers help balance the cost of legacy integration?

Semantic Layer Summit Speakers

1. Dave Mariani. CTO and cofounder of AtScale

2. Josh Klahr. VP of Product for Snowflake's core platform & analytics

3. Bruno Aziza. VP of Data, BI, and AI, IBM

4. Ken Wong: Senior Director of Product, Databricks data warehousing and analytics

5. Francois Lopitaux, SVP and GM of Emerging Tech and Products, ThoughtSpot

6. Chris Lynch, CEO, AtScale

I’ll be writing about what these folks, and other shared - I‘m Mark Palmer, advisor and analyst for Warburg Pincus, Atscale, Dataiku, KX, Data Visualization Soceity, AI.One and Workfusion, and I track the data, analytics, and AI market.

For More on Semantic Layers

New to semantic layers? Read this or listen to episode #4 of the Techno Sapien podcast, Semantic Data 101):

And follow these Substacks and blogs:

• by Try: BI by Another Name

• by

• Eugene Yan. Try: Evaluating the Effectiveness of LLM-Evaluators

• Locally Optimistic. Try The Fallacy of Data-Driven Strategy

• Techno Sapien by Try “The Great Semantic Layer Debate”

Based on this article:

Making such a dramatic claim about one of the most important jobs on the planet —thinking —is heartbreaking. It’s also exciting, because thinking isn’t going away, it’s just changing — shaken to its very core.

For context, I’m passionate about analyst work for several reasons: I was the general manager of a pioneering business intelligence (BI) tool, Spotfire (below), the kind of tool millions of analysts use every day. I serve as an advisor for the Data Visualization Society (DVS), one of the largest networks of visual storytellers. Finally, I’m an analyst, so I get paid to think.

I’ll be speaking at this year’s DVS Outlier conference — my presentation is titled “RIP, BI: The Death and Rebirth of Thinking.” Although that sounds like a dark take, I’m decidedly optimistic about the future of analysts.

Here’s what I’m going to say.

What is an Analyst, Anyway?

Before exploring the demise and rebirth of a billion-dollar industry, millions of jobs, and thinking itself, let’s start with what analysts do and what's at stake.

Analysts are truth sleuths. They formulate questions, set context, then collect, process, and analyze data to uncover insights.2 Then, they craft stories about what they found. If they did a good job, they inspire others to act, or, at least, view the world through a lens with a new tint.

AI fundamentally improves what analysts do, and not just by a little bit — it can help you think between 480 and 6,000 times faster and better.

Those numbers sound hyperbolic? Let me explain my math with a story.

Deep AI Analysis in the TSA Precheck Line at LaGuardia Airport

I was flying out of LaGuardia Airport. During my cab ride from Manhattan, I developed a 700-word prompt for a product idea. I got into the TSA Precheck line and let my prompt fly.

As I shuffled through the serpentine security line, AI went to work. Before going to work, the AI reasoning model I used (Manus) asked questions to clarify my intent. After some back-and-forth, it dove deep. It “read” five books, found and analyzed patents, grabbed quotes from related blog posts, and formulated a detailed report, all before I reached my gate (which, in the new LaGuardia, takes just a few minutes!)

I didn’t have much time before my flight, so I asked AI to flip its findings into six new forms:

1. A long, detailed summary, including a SWOT3 analysis, with links and patent references,

2. A short 1-page executive summary, written in the style of William Zinsser (one of my favorite writers who advocates concise, direct language),

3. A list of 20 non-obvious, counterintuitive ideas AI uncovered,

4. 20 Ernest Hemingway six-word stories that summarized the big ideas,

5. A haiku,

6. A NotebookLM-generated podcast, based on the detailed report.

I downloaded these to my laptop and boarded my two-hour flight.

The results, as anyone who uses the latest AI reasoning models knows, were stunning. The multiple transmission modes of ideas sparked me to find new insights, aha moments, and new questions.

I sketched some ideas in my paper journal (yup—good old-fashioned paper and a multi-color erasable Frixion pen), wrote a few additional prompts for AI, and closed my computer.

I took a nap.

Insights After a Nap: Thinking, 480X Faster and Better

I woke up with a sudden insight: in three hours (including the nap), AI and I had already produced research and seeds of ideas that would have taken months to accomplish on my own, including three truly novel ideas. I did the rough math: my AI-aided thinking process was 480 times faster (three hours versus two months) and, qualitatively, better than I could have done on my own.

The most novel findings, as it turned out, were hallucinations. AI had “synthesized” three ideas from two patents and a blog post — it made them up.

Some fear AI hallucinations; I devour them. I read the sources and found one idea to be fascinating, including one PhD paper that I struggled to understand — I had to ask AI to reprocess and reinterpret it. Ironically, AI’s most valuable ideas come from creative "mistakes"—connections that AI synthesized. Just like humans.

Again, it struck me: AI far outstripped my ability to connect these hidden dots to paint a new, novel picture.

Telling the Story

The final work of an analyst is storytelling, which is how we make an impact on the world, by changing minds.

Again, I partnered with AI. I prompted AI to convey my final ideas to three audiences: a CEO, an engineer, and a board member. For the CEO, that meant a concise executive summary. For the engineer, the technical details. For the board, a strategic view of how the ideas fit into the broader market, impact competitive differentiation, and associated risks. I used Midjourney to create imagery that explained key ideas with visual analogies. I generated over 200 headline ideas and landed on one that I liked.

After about five hours of editing, the result was professional and share-worthy. The final reports were entirely my own, supercharged by AI.

That’s when the analogy of an AI-powered analyst leaping over an old-school BI user came to mind. I asked Midjourney to generate an image of an analyst jumping over a rival analyst, sitting at her desk, crafting data dashboards by hand like a farmer with a hoe in the 1800s:

Why AI is a John Deere Tractor for Analysts

My LaGuardia experience wasn't just a personal anecdote—it revealed three fundamental principles about the transformative power of AI. First, AI demolishes the barrier between the question-asker and the answer. As Einstein said, questions are often more important than answers, and AI transfers the power of question-asking to domain experts.

Second, AI is overwhelmingly more powerful than we are in “dot connection.” For example, the latest AI models are 6,000 times more potent than using Google for conducting manual research. (6,000:1 is the ratio of responses to redirects performed by Anthropic. Read this4, this5, and listen to this6 for an explanation of this idea.) And AI doesn’t tire, take vacations, is cheap, and can simultaneously approach a problem from many angles.

Finally, AI supercharges the art of storytelling—reports targeting multiple personas, using new artwork, and poems. Even hearing your ideas, though jaw-droppingly awesome AI-generated podcasts via NotebookML (Want to check it out? Listen to this AI-generated podcast here or anywhere you listen to podcasts. Literally, all I did was load the text of this article, give it a simple 30-word prompt, and publish it to Substack.)

But what makes AI so valuable creates an existential paradox. The insight gatekeeper role, however well-intentioned and skillfully executed, is fundamentally reordered. This isn't a threat to intelligence itself, but rather a profound shift in where, how, and when that intelligence is revealed, accessed, and applied, and to whom.

Six Steps to Evolve as an AI Age Thinker

Analysts who fail to adopt a new critical thinking approach put their passion to make a difference at risk. This disruption, however, isn't merely a threat to be survived—it's a once-in-a-generation opportunity for transformation.

Here are six ideas to help you not just survive, but thrive as an AI age thinker.

Ask Yourself: “What Makes Me a Special Snowflake?”

Self-knowledge first: Know thyself. Knowing what makes you special forms the foundation for all other steps. Without understanding your unique value, you can’t select tools, craft prompts, judge AI outputs, or create an AI style. Just as being able to play a trumpet doesn’t make you Miles Davis, AI is of little use if you don’t use it to express what makes you special.

So the first step to becoming an AI-savvy thinker has nothing to do with AI. Simply take time to reflect on what makes you truly unique. What wrongs do you want to right in the world? What do you have to say?

What makes me unique? I’m obsessed with how technology meets what it means to be human— how it augments (and, at times, replaces) human cognitive abilities. I’m a tech geek steeped in neuroscience and behavioral economics.

My geek-meets-tech POV helps me pose novel questions to AI and filter its output in ways others might miss.

Got what makes you special in mind? Now let’s get to the AI.

Create a Personal AI Backpack of Tools

Understanding your snowflakeness is essential, but it's not enough—many AI tools obscure rather than amplify what makes you special. The good (and daunting) news is that there are thousands to choose from.

Industry analysts7 8 9 10 estimate that over 10,000 rounds of financing went to AI companies in the last two years. That means there are thousands of AI trumpets for you to play (Miles Davis used just three11!)

Finding AI tools that fit you is a task in and of itself. Test new tools regularly. Curate the ones you like best. As you do, you fill your backpack and build AI muscle. I try a few a month. One of my favorite ways to test AI tools is to try the same prompt on multiple LLMs. You’ll see vastly different results and learn how they work.

Eventually, you’ll cultivate a personal portfolio of tools that align with your unique perspective, workflow, and style of thinking. Currently, my list includes Manus, Perplexity, Midjourney, Otter AI, RunwayML, Grammarly, Dust, and Cursor, but it changes frequently.

Find Your AI Style (Your $20,000 Prompt)

Knowing your tools isn't enough—there's a qualitative leap from playing notes to playing them with artistry. As you work with tools, your AI style will emerge.

After three years, my AI style includes rewriting stories in the style of Haruki Murakami to uncover new analogies and metaphors; using the “But Therefore” rule used by Matt Stone and Trey Parker, the creators of South Park, to check story arc transitions; reimagining visual metaphors in the style of Joan Miró, Paul Rand, David Hockney, Da Vinci, or Jean-Michel Basquiat (The Midjourney-generated image above came from a simple prompt, “/imagine a trumpet in the style of Jean-Michel Basquiat.”)

On LinkedIn, Hamna Aslam Kahn shared this12 prompt from a business consultant that she calls “The $20,000 Prompt” because it embeds the wisdom of a $20,000 consultant, with 41 parts:

If you continually push AI with new angles and prompts, you’ll eventually develop a signature style that produces results only you could have envisioned. Maybe your prompts will be “worth $20,000.”

Edit, Viciously (Don’t Be an Idea Mule)

Kahn has it just half right: a $20,000, 41-part prompt is only valuable if you know how to handle its output. So, AI’s overwhelming “thinking” power creates a paradox: You must edit. Viciously.

Unfortunately, AI is so good that most people skip the best part — the thinking part. Research at MIT found that 68% of participants submitted AI’s output without modification in response to a complicated critical thinking task. That’s not thinking, that’s just carrying AI on your back like a mule (apologies to mules—they’re actually very smart13), adding no value.

Never, ever use or share the output of AI without transforming it into original thinking, adding your unique elements, and taking pride in your work, not the machine’s. I lost count, but I’d hazard a guess that I prompted five different AI tools over 500 times while writing this article, but every word is mine. Every. Single. Word.

So, as William Zinsser said, edit like your readers have better things to do. They can give ChatGPT a banal and boring prompt, too. Put in the work.

Employ Multimodal Thinking

How can you use AI to edit more “viciously?” Think multimodal.

In The Creative Act, Rick Rubin (he produced Johnny Cash, Adele, Tom Petty, The Beastie Boys, and many, many more great artists) wrote: "To vary your inspiration, vary your inputs. Turn the sound off to watch a film, listen to the same song on repeat, read only the first word of each sentence in a short story, or arrange stones by size or color. Break habits. Look for differences. Notice connections." AI helps you work like Rubin in new, fun, analytical, and poetic ways.

I call this “multimodal thinking:” reconsidering ideas through multiple forms, formats, and even sensory channels. My LaGuardia experience showcased six modes of AI-generated input: a deep research report, a poem, six-word stories, my physical paper journal, and a podcast (auditory).

Then, as Rubin suggests, be an antenna: “We are all antennae for creative thought. Some transmissions come on strong, others are more faint. If your antenna isn't sensitively tuned, you're likely to lose the data in the noise. Particularly since the signals coming through are often more subtle than the content we collect through sensory awareness. They are energetic more than tactile, intuitively perceived more than consciously recorded.”

So, think, feel, and listen in a multimodal fashion.

Become an AI Ethnographer

Multimodal thinking alone can lead to scattered insights without a deeper framework for understanding. So, finally, become an AI ethnographer. An ethnographer is a social scientist who immerses themselves in a community to study and understand its culture, behaviors, and social interactions. Do this with AI.

In a way, AI is like a camera14. The better you understand your lenses, lighting, aperture, shutter speed, and their numerous settings, the more effectively you can express yourself.

But AI has infinite settings. So think deeply about how they work and build a second brain with this new, voluminous input. When AI generates a novel idea, have it cite its sources and read them. Journal about them — writing, as Isaac Asimov said, is thinking with your fingers. Using AI’s output without digestion isn’t cheating the world; it’s cheating yourself.

Developing a nuanced, ground-level understanding of AI helps you not only think better yourself, but also become more valuable as an analyst: a better team member, design thinker, and advocate for approaches that leverage machine intelligence and human ingenuity. This isn't just about using AI; it's about shaping its ethical and practical integration and use as an analytical tool by all.

It’s Not RIP, BI: It’s the Evolution of Thought

AI hasn't just arrived; it's a revolution that’s dismantling old ways of accessing and understanding information. Yes, it’s an extinction event for banal analyst work, much like the replacement of the garden hoe for large-scale farming.

But despite this existential threat, something unexpected emerges: Those who embrace what makes them a special snowflake, employ multimodal thinking, and supercharge their signature style will discover that AI doesn’t just alter their jobs—it elevates them, removing routine drudgery and unlocking space for more creative, strategic, and fulfilling work that matters.

So, RIP doesn’t stand for “Rest in Peace,” it’s a new call to evolve and “Rise Into a New Place” — a transformative thinking force that can make you 480 to 6,000 times better, faster, and, ultimately, have more fun.

Come to Miami from June 11th to the 13th for the Outlier conference — I’ll be there, talking about the death and rebirth of analysts, and thinking.

9

Silicon Sands News - State of the Venture Market: 2024 AI Review

Silicon Sands News

State of the Venture Market: 2024 AI Review

Welcome to Silicon Sands News, read across all 50 states in the US and 111 countries…

Read more

a year ago · 2 likes · Dr Seth Dobrin

But, first, let’s back up: What is a semantic layer, and why are they essential to an effective AI strategy?

Wait: What’s a Semantic Layer, Again?

Imagine you're in a foreign country where dozens of languages are spoken. You need to communicate with locals, access services, and understand important information, but you’re at a standstill. This is the challenge that modern AI agents face with enterprise data: databases, applications, tools, and custom code all "speak" different languages.

A semantic data layer acts as a universal translator in this multilingual data landscape. It sits between raw data, stored in various systems and formats, and AI agents — the strangers in a strange land. Just as a translator converts foreign languages into words you can understand, semantic layers transform technical data into AI-understandable concepts and terminology.

In business, without a semantic layer, AI agents can’t understand that a "customer" in one system is the same as a "client" in another, that "revenue" has a specific business definition, and that certain data should be treated as confidential.

Think of it as the difference between looking at sheet music versus an orchestra playing it. Raw data is the sheet music of a business—technically precise but requiring specialized knowledge to interpret. A semantic layer is like the orchestra that transforms code on paper into beautiful music that anyone can appreciate, regardless of their technical background or skill.

It’s not just an analogy: research shows that semantic layers make AI responses three times more accurate than direct database manipulation, and that’s a big deal.

Important Semantic Layer Issues for the AI Age

The opportunity isn’t lost on tech entrepreneurs. Since they’re viewed as the missing link between enterprise data and AI, Semantic Layers were the darling of Gartner’s Data Analytics Summit in Orlando last month, and there are as many ways to express data semantics as there are musical genres.

Here are five essential issues that will be tackled by leaders from Databricks, Snowflake, IBM, and, of course, the host of the Semantic Layer Summit, pioneer in semantic data, AtScale.

The Battle for Semantic Standardization

With numerous approaches to semantic understanding, can the industry converge on open standards that enable AI agents to access and interpret data across organizational boundaries seamlessly? Or, will proprietary semantic models lead to new forms of confusion and vendor lock-in? What is the role of Anthropic’s Model Context Protocol, or MCP?

Dael Williamson, EMEA Field CTO at Databricks, addresses these questions in his session, "Open Semantics: AI and Interoperability." For technical leaders, this debate has profound implications. AI agents require consistent, contextual understanding of data to operate effectively across systems. Without standardized semantic layers and programming interfaces, organizations risk creating isolated pools of intelligence—essentially replicating the data silos of the past in a new, more sophisticated form.

Natural Language Interfaces: Promise vs. Reality

"For over a decade, business users have been told this-or-that tool would take them to the promised land of Self Service Reporting," Joseph Hobbs from Databricks wonders: Will natural language finally deliver on the promise of democratized data access, and what role will human analysts play?

Natural language querying (NLQ) interprets human instructions and translates them into precise data operations. But skepticism remains: Will NLP truly empower business users or create new forms of technical debt that IT departments must manage?

The Automation Dilemma in Data Integration

"Driving Innovation with a GenAI-Powered Smart Data Fabric and Universal Semantic Layer" addresses a third controversial area: How to balance automation and human oversight.

Joe Lichtenberg of InterSystems will suggest that smart data fabrics provide the unified foundation that agents need to operate across organizational silos. However, the spread of AI agents raises new questions about governance and control. How much autonomy should agents have in interpreting and integrating data? What safeguards are necessary to prevent semantic drift and misinterpretation?

The Future of Human Analysts

Will GenAI ultimately replace traditional business intelligence (BI) analysts or merely augment their capabilities? The panel on "The Future of Semantics and BI with GenAI," featuring leaders from Snowflake, IBM, Databricks, and ThoughtSpot, explores this question.

It’s a tricky transformation to tackle. AI agents accelerate insight generation and decision-making, but also alter the nature of analytical work, risking the loss of creativity, instinct, and context that humans bring to critical business decisions.

The Real-World Battle Between Cost and Complexity

Multiple sessions address the practical challenges of modernizing legacy analytics systems. Robert Skerry and Pedro Alexandre Simão from Celfocus explain how Vodafone Portugal balances the costs of modernization with the weight of maintaining legacy systems. And Adam Walker from Telus explores how a semantic layer helped simplify its network analytics by abstracting away vendor-specific complexity in "Scaling Telecom Analytics with a Semantic Layer."

These real-world illustrate the implications of AI agent ecosystems on existing systems, and how semantic layers help fit old school thinking with new, innovative AI systems.

Join the Conversation! Navigating the Semantic Layer / AI Frontier

As AI agents become more central to business operations, the semantic layer that enables them to understand and interact with organizational data grows increasingly critical. The AtScale Semantic Layer Summit 2025 will bring these controversies into sharp focus, challenging business and technical leaders to rethink their data strategies for an agentic future.

For decision-makers attending the summit, the key will be balancing pragmatic near-term implementations with longer-term strategic considerations about interoperability, governance, and the evolving relationship between human and artificial intelligence. Those who successfully navigate these controversial waters will position their organizations to harness the full potential of AI agents and agentic applications in the years ahead.

To follow my takeaways from the Semantic Layer Conference, follow me on LinkedIn or subscribe to !

Using AI isn’t artificial if you do it right. It’s a tool that enhances your ability to capture and express emotions. Your stories, your feelings are the ink that AI uses to draft beautiful prose; it’s your job to make it yours.

Unfortunately, 68% of us cheat with AI.

Here’s how to do it right.

Prompting Your Way to a Personal Poem

In a research study at MIT,1 68% of human participants passed off AI’s work as their own, without modification. This isn’t just cheating; it misses the power of AI. A better way to use AI is like a camera: you decide where to point it, when to press the shutter, how to adjust the scene, and which images to keep.

To write a poem, start by prompting AI with your raw emotions, your “scene.” That is, don’t ask AI to “Write a poem about my daughter”; take some time and think.

For my daughter’s college graduation, I spent a few days reflecting on what makes my daughter special. For Christmas, we both got Snoopy and Woodstock tattoos, so I wrote down, “We’re like Woodstock and Snoopy, on an adventure together.” She loves old people. We both love One Direction. She knits while watching TV.

These ideas are like subjects in a photograph - it’s the raw material.

AI is the Ultimate What-If Engine

Think of AI as a “What If” engine for creative work. These What-If scenarios are like setting the scene for a camera, except you can set it to anything you dream of. You invent the scene, the lighting, and the concepts with your AI prompts.

For example, I fed AI scenes like:

• What-If the poem used Woodstock as my daughter and Snoopy as me?

• What-If the poem expressed the idea of giving a child “roots and wings.”

• What-If the poem was about her graduating from High Point University?

AI had some amazing ideas, but that’s where the work starts. I kept, deleted, rewrote, and reorganized its ideas, including this stanza:

AI can generate hundreds of ideas based on your What-If prompts. I prompted it to write jokes, simplify ideas, use new metaphors, use different styles, or find new rhymes. AI is an active, eager thesaurus, powered by your prompts.

You’re the editor. Like discarding most of your bad pictures, discard most of AI’s ideas until you find the perfect words.

Mixing Modes: What-If We Wrote a Haiku? What-If We Made it Funnier?

AI and I tried a Haiku (after about 10 prompts and passes at editing, I copied my draft into AI and thought: “What-If this was a Haiku?”

Here’s what I got:

The depth of the What-If questions you can ask is infinite. Keep asking, keep prompting, keep editing.

This is a healthy and fun way to use AI.

Is it Artificial?

Using AI isn’t artificial if you use it right.

First, choose your ingredients. Like taking pictures with a camera, you choose your subjects, settings, lighting, and scene.

Treat AI as a “What-If” machine. The quality of your work depends on the quality of your questions.

Always rewrite. Aggressively evaluate AI’s suggestions, discard most, tweak what’s close, and stitch together the pieces that capture your meaning and emotion.

The final piece is all yours, born from your heart, crafted with a 21st-century quill.

Like this? Listen to it as a NotebookLM-generated podcast:

Read my favorite Substacks:

1. by Michael Spencer

2. Marcus on AI by Gary Marcus

3. Last Week in AI

4. The Algorithmic Bridge by Alberto Romero

5. The Intrinsic Perspective by Erik Hoel

6. The Pragmatic Engineer by Gergely Orosz

7. Kareem Abdul-Jabbar

8. Polymathic Being by Michael Woudenberg

9. The Muse by Birgitte Rasine

10. The Creative Act: Thoughtforms & Innerworks by Rick Rubin

This episode of the Techno Sapien podcast likens using AI to operating a camera, where the user controls the input and selects the best outputs, and also to a "What-If" engine for generating various creative ideas based on prompts. Ultimately, the text stresses that the user must act as the editor, refining and personalizing the AI-generated content to create a truly original final piece.

Read the original article on the Techno Sapien Substack, by Mark Palmer.