In June, Pew asked Americans how they feel about AI in daily life. 52% said they were more concerned than excited about AI in daily life. Only 9% said more excited, the lowest since Pew started asking in 2021.
A month earlier, Gallup asked American workers how often they use it. 52% said they use it at work at least a few times a year. 15% use it at work every day, up from 4% in 2023.
Same country, a month apart, same 52%. The slogan says everyone hates AI and everyone uses it. The numbers say about half worry and about half use it at work. Two separate surveys can’t tell you how many are the same people. My guess is a lot.
The Quality Fight: Boring and Real
Sonar surveyed 1,149 developers for its State of Code report. Only 4% fully trust that AI-generated code is functionally correct. 61% say it often looks correct but isn’t reliable. And only 48% fully agree that they always check it before they commit.
So almost nobody fully trusts the code, and fewer than half will say, without hedging, that they always check it.
I’ve supervised thousands of AI coding sessions across my teams. When my teams review the output and enforce the specs, very little breaks. The tool mostly does what it’s told. The problem is that checking takes effort, and it can eat exactly the effort people hoped AI would save. That makes the first plausible result tempting.
When people skip that review, bad output gets through. Spec files get ignored. Deprecated APIs get suggested with confidence. It writes bad code. I’ve tested that on my own work. As I’ve written before, the worse part is that the bad code looks plausible, and plausible is more dangerous than obviously wrong because it invites you to stop checking.
In my experience, only a minority can point at a stack trace or an incident ticket and say, “Here. This is what I mean.” The distrust is much broader than the vocabulary to describe it. But most of the yelling isn’t about code quality.
The Fear Fight: Pointed at the Wrong Throat
People are losing sleep over whether their role still exists in three years. I’m one of them. Frontend-only roles are folding into full-stack: between January 2024 and January 2025, front-end postings fell 25% globally while full-stack postings rose 9%. In the UK, permanent jobs citing PHP dropped from 918 to 404 in a year. Not all of that is AI, and tech hiring was falling before the models got good. I run teams that ship with AI every day, and I still wonder what this industry looks like in three years and whether my own seat is in it.
Stanford’s 2026 AI Index put experts and the public side by side on the same question. 73% of AI experts think AI will improve how people do their jobs. 23% of the US public agrees. It’s not just about who’s read more papers. It’s also about who holds equity and who holds a cardboard box under their desk.
I’ve covered the walk-backs before: the bloodbath predictions, “I need less heads,” Klarna bringing humans back.
Challenger, Gray & Christmas tracks announced US layoffs and the reason given for each. Challenger says AI dominated the reasons through spring and summer. By the end of October, employers had cited it for 121,401 cuts this year, about one in five. In October alone it was cited for 1,265, the lowest monthly total of 2026, and market and economic conditions overtook it as the leading reason for the year.
My reading is that AI was the reason that sounded best in a press release this spring, and the economy sounds better this fall. Maybe the AI wave just ran its course. I can’t rule that out. But the reason in a layoff announcement is a sentence somebody picks in a meeting, and the model isn’t in that meeting.
To be fair, total announced cuts through October are down 44% from last year, the lowest since 2022, and announced hiring plans are up 34%, most of them seasonal. AI-blamed cuts piled up in a year when announced layoffs overall fell.
In Robert Half data reported by CNBC in July, 32% of US hiring managers said they had cut a role mainly because of AI and later rehired for the same or a similar one. Separately, Forrester predicted last year that half of AI-attributed layoffs would be quietly reversed, with the jobs coming back offshore or at lower pay.
If people sound unhinged about AI, part of the reason is that the announcement is loud and the rehire is quiet.
Data centers have become a fight of their own. Local opposition blocked or delayed at least 75 data center projects worth about $130 billion in the first quarter alone, the most since Data Center Watch started tracking in 2023. On November 3, voters backed restrictions or rejected pro-development measures in at least 37 of 41 local contests, according to Trepp’s unofficial tally. Power bills and water are local worries, and voting on them is aimed exactly right. If the argument is about layoffs, though, the server rack is the wrong target. That decision belongs to the people who ordered them.
Incognito Mode
In a 2025 KPMG and University of Melbourne study across 47 countries, 57% of employees who use AI at work admitted to using it without being transparent about it, by not disclosing it or by passing its output off as their own. Sonar found 35% of developers reach the leading AI tools through personal accounts. In BlackFog’s shadow AI survey, 60% of 2,000 UK and US employees at companies with more than 500 staff said unsanctioned AI is worth the security risk if it helps them work faster or hit a deadline.
Picture someone who clicks “more concerned than excited” on a survey, then opens an incognito tab at 11:47 p.m. and pastes three thousand lines of company code into ChatGPT. No survey links those two answers. I bet you’ve met that person anyway. Using a useful tool doesn’t mean trusting what companies will do with it.
Sometimes the evidence leaks into print. Last year readers found an AI editing note left in chapter three of a published romance novel. A syndicated summer reading list that ran in the Chicago Sun-Times recommended books that don’t exist.
Those screenshots are rare, which is why each one goes viral in an hour. Signs of AI writing, on the other hand, are common in Pew’s web sample. Pew ran almost half a million English-language web pages through an AI detector and found significant signs of AI writing on one in ten pages collected in July 2026, and on more than a third of that sample’s pages published after ChatGPT launched. The leftover prompt is the one somebody noticed.
Tell everyone AI is forbidden, then act shocked when people use it in private anyway. A ban doesn’t guarantee anyone stops.
So the quality critics are right. The fear is justified, but the layoffs were announced by executives, with reasons executives picked. And the distrust in the polls sits right next to use people don’t disclose at work.
The Position Nobody Occupies
I expanded our project boilerplates with AI-built functionality: RAG pipelines, agentic flows, Power BI integration, Azure Search. That cut project estimates and lowered costs for clients. I built an internal app for collecting team health data. I’m rewriting a 7-year-old financial system that runs the core of the business. Without AI, I couldn’t have done half of it.
Not a single person was fired to make this happen. I hired five. The boilerplates didn’t replace engineers. They made projects cheaper for clients and faster to start, which meant more projects, which meant more people.
That’s one shop, and everyone in it reviews the output. Review matters. So does what you choose to do with the capacity you gain.
Yes, AI can move the work faster. You can use that to finally ship the boring plumbing your team never had time for, or to cut heads and pray the outages arrive after your bonus vests.
The Linux kernel wrote an AI policy I can live with. AI can help write a patch. Only a human can add the Signed-off-by, and that human takes full responsibility for it.
As I wrote in Announcement Problem, there are two conversations about AI right now. Conference stages and Slack channels. They don’t overlap. I’ve been in the second room for years. The tools help. They do not replace judgment.
Plenty to Hate. Don’t Stop at the Model.
CEOs who predict AI will replace your job, and layoff announcements that blamed AI in the spring and the economy by fall. Memory prices I wrote about in January that kept climbing this year. Residential electricity prices that EIA expects to rise again this year. Power demand climbing with data centers and electrification. Water for cooling in towns now voting against the next data center. Pricing where nobody can tell you what you actually get for $20, $100, or $200 a month. What looks more and more like a bubble. Companies that trained on web content collected without its creators’ permission and now sell access to the resulting models.
The models are messy. They hallucinate, leak, discriminate. Engineers should be losing sleep over that. But the screaming gets louder when systems with those bugs are sold as the reason five hundred people can go.
I spend my days using AI to build systems that make my team more effective and my clients’ projects cheaper. Nobody got fired for that.
Hate it in the poll if you want. Use it at 11:47 p.m. if you have to. But if you need someone to be angry at, start with whoever picks the reason in the layoff announcement. Across the announcements, this spring it was AI. By October it was the economy.