Last updated: August 19, 2026 · By Vishal Swami, Founder & Lead AI Reviewer, AISagely
"My friends all hate AI; I just joined an AI startup" isn't a hypothetical — it's the title of a real essay Rachel Thomas, co-founder of fast.ai, published on August 18, 2026, after friends started forwarding her anti-AI takedowns as if she'd never seen them. If you're in the same spot, working in AI while the people around you have soured on it, the useful move isn't picking a side between "AI is all hype" and "the critics don't get it." It's going through the backlash complaint by complaint, keeping the parts that are documented and correct, and dropping the parts that are just noise. That's a slower conversation than either extreme, but it's the only one that survives a follow-up question.
Short answer: Rachel Thomas of fast.ai wrote about facing exactly this in an August 18, 2026 essay: friends citing real harms — AI-generated spam, scams, and a $3 million LAUSD chatbot whose founder was charged with fraud — while she kept building. Her answer wasn't to defend AI wholesale. It was to admit the harms are real, then build tools that keep humans in control instead of replacing their judgment.
Where this actually comes from
Thomas didn't write a defensive rebuttal. She listed her friends' complaints without softening them: the internet is filling up with AI-written filler, execs make claims so overhyped they "verge on fraud," professors are rewriting curricula around AI-generated essays, and open-source maintainers are getting flooded with low-quality AI-generated pull requests. She also named the ugliest example she had — Los Angeles Unified School District paid $3 million toward a $6 million contract for an AI chatbot called "Ed" before its maker, AllHere, collapsed and founder Joanna Smith-Griffin was charged with securities fraud, wire fraud, and identity theft.
Then she made her case anyway: "There is a genuinely useful technology underneath those false and over-hyped claims about what AI can do." She'd rejoined Answer.AI, the company that grew out of fast.ai, to work on SolveIt — a tool built so people edit the AI's output directly instead of trusting it blind. Her line for why: "I want to use AI in ways that don't degrade my own critical thinking skills." She closes on a point worth sitting with: "Their values and decisions aren't the only version of what AI technology can be" — meaning the loudest, most overhyped companies don't get to define the whole category for everyone building in it.
None of this is happening in a vacuum, either. The same week Thomas's essay went up, Futurism reported on a CNBC Generation Labs poll finding that a majority of young US adults said they don't trust a single sitting AI executive to act responsibly — not one of the nine names asked about cleared a positive trust score. Your friends' skepticism isn't an outlier opinion right now. It's closer to the median one.
Step-by-step: what to actually do when your friends hate your AI job
I test AI tools for a living, so I get a version of this conversation constantly. In my testing of dozens of AI products for this site, the pattern I keep running into is that the harshest critics are almost never wrong about the specific thing they saw — a bad chatbot, an inflated benchmark, a scraped blog post. They're wrong about how far that one example generalizes. Here's the process that actually works, not the one that just wins the argument.
1. Write down the specific complaint, not the vibe
"AI is bad" isn't a claim you can respond to. "AI-generated PRs are burning out open-source maintainers" is. Ask your friend for the specific thing that set them off — a story, a headline, a bad experience at work — before you say anything back.
2. Check whether the complaint is true
Most of the complaints Thomas's friends raised check out: students really are submitting AI-written essays, maintainers really are drowning in low-effort AI PRs, and the LAUSD chatbot really did collapse into a fraud case. Don't argue with facts that are correct. My breakdown of what’s actually happening to jobs is a good model for this — some of the fear is earned, some isn't, and the two need different responses.
3. Separate the harm from the technology
A scam built with AI is a scam. A chatbot company that lied to investors is fraud. Neither indicts every tool that uses a language model, any more than a bad hire indicts your whole industry. This is the distinction Thomas leans on — the harms are real, but they're not the only thing AI can be.
4. Show them what you're actually building, not what the industry is selling
Vague reassurance ("it's not like that where I work") doesn't land. Specifics do. Thomas points to SolveIt's design — the user edits the AI's response instead of accepting it — as the concrete difference between her work and the products her friends are angry about.
5. Let the disagreement stand where it should
You don't need your friends to love AI. You need them to know you've actually reckoned with their specific objection instead of deflecting it. That's a lower bar, and it's usually the one that ends the argument instead of restarting it.
Example prompts you can copy
Use these before a hard conversation, or to pressure-test your own reasoning first:
- "Here's a friend's specific complaint about AI: [paste it]. Tell me which parts are documented fact, which are speculation, and which are opinion dressed up as fact."
- "I work on [your actual product]. List the strongest, most specific objection a critic could make against it — not a generic AI complaint, one that applies to this product."
- "Rewrite this explanation of my job so it names one real limitation of the tool, instead of only listing benefits."
The third prompt is the one that changes conversations. A pitch with zero downsides reads as marketing, and your friends can tell.
Common mistakes to avoid
The mistake I see most is treating every AI critique as the same critique, then rebutting all of them with the same one good example. It doesn't work — a friend worried about AI-generated scams isn't reassured by a story about a great coding tool. I've also watched people over-correct the other way and agree with every complaint just to end the conversation, which quietly reinforces the idea that there's nothing worth defending, and that's not honest either. Last one: don't lead with the industry's talking points. Thomas didn't. She led with her friends' worst examples and answered those directly, which is a harder conversation but the only one that actually holds up.
A fourth mistake, smaller but common: assuming the conversation only goes one direction. When I tested this with my own circle of friends outside tech, the ones who pushed back hardest usually had a specific, recent story behind it — a scam text, a school policy, a coworker leaning on a chatbot instead of doing the work. Asking what happened, instead of assuming you already know, changes the whole conversation from a debate into an actual exchange of information.
The complaint versus the evidence
| Common complaint | Is it accurate? | What actually backs it |
|---|---|---|
| "AI execs overhype what their products can do" | Often, yes | A 2026 CNBC Generation Labs poll found young adults distrust every major AI executive surveyed — see young people hate AI CEOs |
| "Schools are wasting money on AI scams" | Yes, documented | LAUSD paid $3M of a $6M contract for the "Ed" chatbot before its founder was charged with fraud |
| "Open-source is being flooded with AI-generated junk" | Yes, widely reported by maintainers | Cited directly in Thomas's fast.ai essay from her own network of maintainer friends |
| "AI is going to replace most jobs soon" | Overstated | My analysis of the AI jobs apocalypse finds the displacement story is messier than the headlines |
| "Every AI product is the same" | False | Design choices vary a lot — see how working with AI differs from just automating tasks |
Tools that make this easier
If you want to have this conversation with evidence instead of vibes, test the actual product before you defend it. My AI tool reviews explain how I test these things instead of repeating a vendor's pitch, and AI tool ratings walks through how to read any review, mine included, without getting fooled by a cherry-picked demo. If your friends' real objection is "which of these tools is even worth using," Best AI Models (2026) is the ranked list I currently stand behind.
My take
Thomas's essay works because she doesn't ask her friends to change their minds about AI in general. She agrees with the specific, documented harms and then explains what she's building instead. That's the model I'd copy: don't defend the industry, defend the thing you actually made, and be honest when you can't.
Frequently Asked Questions
Is it true that Rachel Thomas's friends all hate AI?
Yes — in her August 18, 2026 essay on fast.ai, she describes friends sending her anti-AI articles and citing real problems: AI-generated spam, overhyped executive claims, and the LAUSD chatbot fraud case, among others.
Why do so many people hate AI right now in 2026?
A mix of real harms and real overhype. Scams, low-quality AI content, and failed AI contracts like LAUSD's are documented. A CNBC Generation Labs poll also found deep distrust of AI executives specifically, which is a different problem than distrust of the technology itself.
What is SolveIt, the tool Rachel Thomas works on?
SolveIt is a tool from Answer.AI, the company that grew out of fast.ai, built around George Pólya's problem-solving framework. Instead of accepting an AI's output as final, the user edits it directly and decides what happens next.
How do I explain my AI job to friends who are skeptical?
Address their specific complaint, not "AI" as a category. Confirm what's actually true about it, then explain the concrete design choices in what you're building that differ from the thing they're angry about.