Something odd sits at the center of the AI boom: the people shipping products at record speed have rarely seemed further from the people actually using them. AI executives talk about a transformation everyone is supposedly welcome to, while public polls grow more anxious every quarter. The distance between how insiders experience this moment and how everyone else does is now a story in its own right — told through survey data, product walk-backs and meme-worthy executive tweets. Here is what the numbers say, where the gap comes from, and what it would take for AI executives to close it before the gap closes the market for them.
- The Widening Gap Between Insiders and Everyone Else
- What Ordinary People Actually Think of AI
- What AI Executives Keep Misreading About Their Own Products
- When the Executives Themselves Speak
- Who Builds AI, and Who It Gets Built For
- The Pushback Is Working — and Being Answered
- What Building for the General Public Actually Requires
- Frequently Asked Questions
- References
The Widening Gap Between Insiders and Everyone Else
What the Stanford report actually measured
The clearest measurement of the divide comes from Stanford’s 2026 AI Index Report, which documented something critics had long suspected: AI insiders and the general public now live in different informational worlds. Asked whether AI would have a positive impact over the next twenty years, 84% of experts expected better medical care; only 44% of Americans did. On jobs, 73% of experts saw AI helping versus 23% of the public, and 69% expected a positive economic impact against just 21% of everyone else.
Only 10% of Americans were more excited than concerned, and 64% expected AI to mean fewer jobs — numbers that put the priorities of AI executives in sharp relief. As TechCrunch summarized the findings, AI leaders have been consumed with managing advanced systems, while “everyday folks are more concerned about AI’s impact on their paycheck and whether or not their power bills will go up.”
Why AI executives miss the divergence
The gap is not conspiracy; it is geometry. AI executives work inside the bubble all day: they hire from the same small pool of universities, raise money from the same funds, speak at the same conferences, and prototype on themselves. Their tools write their code and draft their email, so the revolution feels personal and obviously good. Their feedback loop — investors, power users on X, benchmark charts — filters enthusiasm back to itself. When the sample you listen to is assembled entirely from people who already agree, a public that disagrees starts to look confused rather than informed.
This is not a failure of intelligence or kindness among AI executives; the bubble’s damage is structural, turning unexamined defaults into universal ones.
Where AI executives’ certainty ends and the public’s begins, the Stanford data draws the line with a ruler:
What Ordinary People Actually Think of AI
The numbers have moved in one direction
The polling trend is unambiguous. In June 2026, Pew Research reported that just 16% of Americans expected AI’s impact on society over the next twenty years to be positive, while 40% expected it to be negative. Some 59% did not trust AI companies to develop the technology safely, and 67% did not believe the U.S. government would meaningfully regulate it. Even among Americans under 30 — the demographic the industry treats as native adopters — only 14% expected more benefit than harm. None of this registered as a crisis inside the industry’s own dashboards, where AI executives watch usage curves climb and conclude the public is coming around.
Other surveys point the same way: 52% of Americans were more concerned than excited about increased AI use in daily life, up from 37% in 2021, and more than 70% said AI was advancing too quickly. The numbers are not noise; they reflect a public that has watched AI executives’ promises and kept receipts.
Worry does not mean non-use
The most instructive dataset came from Gallup’s 2026 survey of roughly 1,000 people in each of 37 countries, commissioned by Microsoft. In the United States, 74% of people are worried about AI — and 68% of those who use it every day are worried too. Gallup senior scientist Pablo Diego-Rosell put the nuance plainly: people “can be curious about AI, expect it to be beneficial and use it frequently while still being worried about it.”
That nuance is where AI executives’ playbooks keep failing. Adoption is not enthusiasm. People will use the features they are given while distrusting the people giving them — a distinction that matters when your growth plan assumes usage breeds affection. Abroad the picture brightens — in Singapore 46% use AI daily and more than 80% expect life to improve — which suggests the problem is not AI itself but what local builders have chosen to ship, and for whom.
The demographic split inside the United States is just as stark: chatbot use runs higher among men and among adults under 50, yet product roadmaps rarely show evidence of that difference — as if AI executives had sampled the audience for a retirement-banking product from their own org chart. No poll commissioned in 2026 shows AI executives winning the public’s confidence.
| Survey | Scope | Headline finding |
|---|---|---|
| Pew Research (Jun 2026) | U.S. adults | 16% expect positive impact over 20 years; 40% expect negative |
| Pew trend data (Aug 2026) | U.S. adults | 52% more concerned than excited, up from 37% in 2021 |
| Gallup (Sep 2026) | ~1,000 × 37 countries | 74% of Americans worried; 68% of daily users worried |
| Stanford AI Index (Apr 2026) | U.S. and global | Only 10% of Americans more excited than concerned |
| CNBC (Aug 2026) | Ages 18–34 | Majority distrust named AI-industry leaders to act responsibly |
What AI Executives Keep Misreading About Their Own Products
Hardware that shipped into a shrug
The Copilot key is the canonical cautionary tale. In January 2024, Microsoft gave Windows keyboards a dedicated AI key, described at the time as the first big change to the Windows keyboard in 30 years. Users mostly ignored it, and by late 2026 Microsoft and PC makers were quietly retiring the Copilot+ PC brand after what trade coverage called security concerns, unsuccessful marketing and inconsistent branding.
Windows Recall — the feature that screenshotted your desktop every few seconds — arrived only after a forced delay, following a privacy backlash AI executives had been warned about months in advance. None of these products failed because the technology was weak; they failed because nobody outside the lab was asked whether the idea was wanted at all.
Features pulled within days
The walk-back list grew long enough to need its own maintenance window. Meta shipped chatbots that messaged users first, betting that proactive pings would become companionship. Instagram’s Muse Image let anyone remix any public account’s photos with an @-mention; after users and talent agencies revolted, Meta pulled it within days, conceding the feature “missed the mark.” OpenAI folded its Atlas browser and killed Sora in a retreat from “side quests,” feeding a running AI graveyard list of projects that died without an audience.
Each failure was survivable. The pattern is what should worry AI executives: the features people rejected were precisely the ones the builders used least and understood least.
| Product decision | The bet | The outcome |
|---|---|---|
| Copilot key and Copilot+ PCs | One key revives the PC category | Ignored; brand quietly retired by late 2026 |
| Windows Recall | People want a searchable memory of their PC | Delayed by privacy backlash; shipped constrained |
| Meta’s proactive chatbots | Nudges become relationships become revenue | Read as creepy; internal conduct scandal followed |
| Instagram Muse Image | Everyone wants to remix everyone’s photos | Pulled within days — “missed the mark” |
| OpenAI Atlas and Sora | Browser and video can ride the core brand | Both shut down in July 2026 as side quests |
When the Executives Themselves Speak
The thank-you note that lit up the timeline
In March 2026, OpenAI CEO Sam Altman posted “so much gratitude to people who wrote extremely complex software character-by-character” — a nostalgic nod that landed badly while his industry was shedding the workers it no longer needed, echoing Amazon’s 16,000 cuts and Block’s near-halving of staff. The most-quoted reply of the day was itself a product critique: “Billion dollar app idea: AI that reads billionaire tweets before they post them and says ‘this is going to make you sound incredibly out of touch, are you sure?’”
Moments like this are why the trust gap is emotional, not intellectual. The public does not hate technology; it hears AI executives describing a future everyone will pay for in ways that never feel like they belong to everyone — and increasingly says so out loud.
Manifestos, memes and fines
Mark Zuckerberg published a 6,500-word personal-AI manifesto arguing for a PhD-level tutor for every child and “dynamic auction” pricing for compute. TechCrunch’s verdict was blunt: the essay is exactly why people don’t like AI, calling its examples “alarmingly out of touch with the reality of how AI tools are being used.” That same August, a court fined Meta $567 million over harms to children — context that made the company’s new AI-optimism ad, scored to Bowie’s “Five Years” (a song about the Earth running out of time), land as satire.
At Davos, Microsoft’s Satya Nadella was more candid, warning that if more people were not using AI, “more or less… it’s going to be a bubble and a popped bubble.” He also claimed consumer Copilot’s daily users had grown nearly threefold year over year — without sharing the number. What is known: roughly 15 million paid Microsoft 365 Copilot seats against a potential base of 450 million, about 3%. That is the real number AI executives should be planning around. None of this requires malice or stupidity — only insulation, which corner offices, keynotes and echo-chamber boards are excellently engineered to provide.
Who Builds AI, and Who It Gets Built For
A participation gap, not just a PR gap
At SXSW in March 2026, emotion-AI pioneer Rana el Kaliouby called today’s AI industry a boys’ club and warned that if women are left out of building it, “we’re going to have widened the economic gap like crazy.” The concern is structural: a team that hires in its own image produces a roadmap with its own blind spots, and no amount of marketing later fixes who was in the room when scope was set.
The young are paying attention. When CNBC named nine top figures of the AI industry for people aged 18 to 34, a majority said they did not trust them to act responsibly — a rebuke AI executives can treat as noise, or as a demographic warning about who will be buying (and voting) in ten years.
Designing from your own life
Usage data shows how narrow the builder’s reference class has become. About 25% of U.S. adults use chatbots daily, men at 27% versus women at 20%, and a full 75% of Americans aged 65 and over have never used one — meaning the heaviest users resemble the people who make product decisions, and everyone else is an edge case.
Teens bend the pattern the other way: three in ten use chatbots daily, and Character.AI usage doubled in households earning under $75,000 — the most dependent users are the least represented in the rooms where AI executives choose what ships next.
When builders design from personal experience, they prototype for affluent power users with high tolerance for breakage.
Airbnb CEO Brian Chesky named the fix plainly: “part of it’s a narrative issue… but part of it is we need to actually be developing more products that just regular people can use.” Stanford psychiatrist Nina Vasan added the obligation: people use AI for emotional support whatever the intent, which means “companies do have a responsibility to adjust their models to be solving for user well-being.” Our own tour of the prosumer AI market keeps finding the same seam: the products ordinary people love are built by teams that watched them, not at them.
The Pushback Is Working — and Being Answered
Walk-backs as market signal
Every retreat now carries a price, and the pattern is teaching. Character.AI ended open-ended chat for all under-18 users after lawsuits tied to teen suicides, expected “some churn,” and funded a new safety lab. Leaked guidelines showed Meta’s bots had been allowed romantic chats with children — the kind of disclosure that converts regulators’ patience into subpoenas, and that puts every maker’s privacy promises under watch.
Meanwhile the no-AI counter-market grew: DuckDuckGo’s traffic boomed as it made its AI-free search easier to reach — proof that “just don’t ship the AI” is itself a viable product position, and one that ordinary users are voting for with their attention.
Ordinary users have leverage AI executives can no longer wave away — shipping first and apologizing later is now a measurable cost, not a free option.
Regulators are no longer patient
The political pressure is compounding. Max Tegmark’s Pro-Human AI Declaration, circulating since March 2026, cites polling that 95% of Americans oppose an unregulated race to superintelligence, and its signatories run from Steve Bannon to Susan Rice. California passed a first-in-the-nation AI companion law; federal bills followed. Tegmark’s declaration even asks for mandatory pre-deployment testing of chatbots aimed at children — the exact review most companies skipped voluntarily. The United Nations has also called for urgent action to rein in frontier AI. Our coverage of why Democrats now see AI regulation as a midterm edge explains the domestic arithmetic.
For AI executives, regulation is the out-of-touch tax: every product shipped without ordinary users in the room becomes a law written later without them either.
What Building for the General Public Actually Requires
Deliver the big promises, loudly
Anthropic CEO Dario Amodei diagnosed the gap better than any consultant: “the most accurate criticism of AI companies… is that we haven’t yet delivered on our big promises to benefit the world. That is totally on us.” The fix is not a messaging memo. People forgive a clunky interface; they do not forgive a cancer that did not get cured on schedule.
That means measuring what matters outside the bubble — fewer night-shift pickups, cheaper prescriptions, homework actually explained — and telling those stories in ordinary language. AI executives keep underselling boring wins because benchmarks impress peers more than guardrails do, and that instinct is exactly what the trust gap is made of. AI executives who can name one ordinary user whose week got measurably easier have a story; those who cannot have only benchmarks, and benchmarks have never comforted anyone.
Test with the people you keep skipping
The practitioner’s playbook already exists: Google’s People + AI Guidebook packages human-centered research methods for teams shipping AI, and hands-on evaluations like our test of every major AI agent show what independent scrutiny catches when labs grade themselves. The practical bar: recruit test panels that match the population — people over 65, low-bandwidth users, non-engineers — before a feature reaches hundreds of millions. Teams that skip that step pay twice: once in headlines, once in hearings.
Nadella gestured at the same idea at Davos, arguing AI must be “equitable across all these different communities” rather than “concentrated… only in the wealthy places.” Nobody expects AI executives to abandon ambition. The ask is smaller and harder: ship as if the general public were your user, because they already are.
| Dimension | Out-of-touch shop | Grounded shop |
|---|---|---|
| Success metric | Benchmark leaderboards | Tasks completed by representative users |
| User sample | Staff and power users | Panels matched to age and income |
| Launch posture | Ship, defend, walk back | Test, tune, then ship |
| Trust strategy | Messaging memo after failure | Delivered, visible wins first |
| Who is in the room | AI executives and early adopters | The actual audience |
Frequently Asked Questions
Why are AI executives out of touch?
They live inside a bubble with its own economy: the same universities, the same investors, the same feedback channels, and daily heavy use of the tools they build. The loop confirms enthusiasm back at itself, so a hesitant public reads as confused rather than as informed disagreement. Start with the boring audits: who was in the room when scope was set, and who was never asked at all.
Do people avoid AI because they don’t understand it?
Mostly no. Gallup found 68% of Americans who use AI daily are still worried about it, and Pew’s researchers concluded that consumers understand AI well enough as it is — they just do not think the trade-offs are worth it. The concern is informed, not ignorant, and AI executives keep treating it as a comprehension problem.
Which product decisions best show the gap?
The Copilot key, Windows Recall, Meta’s proactive chatbots, Instagram’s Muse Image — pulled within days of backlash — and Atlas and Sora, both killed as side quests. Each was built for a user the executives resembled, not the user who actually showed up.
What would close the gap fastest?
Delivering the big promises Amodei named, testing with representative users before launch, budgeting for walk-backs, and publishing plain-language wins. None of it requires new technology — mostly it requires that AI executives spend one week inside an ordinary user’s day.
References
Stanford HAI — AI Index 2026 Report
TechCrunch — Stanford report highlights growing disconnect between AI insiders and everyone else
TechCrunch — Only 16 percent of Americans think AI will have a positive impact on society
TechCrunch — Even Americans who use AI every day are worried about it
TechCrunch — AI was supposed to win people over by now — it hasn’t
TechCrunch — Mark Zuckerberg’s AI manifesto is exactly why people don’t like AI
TechCrunch — Meta removes controversial AI feature on Instagram after backlash
TechCrunch — Sam Altman’s thank-you to coders draws the memes
TechCrunch — Character.AI is killing the chatbot experience for minors
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
