Companies are adopting AI wrong, and it's going to cost them

There's a specific kind of meeting happening in companies right now. A senior executive reads an article about how a competitor saved millions using AI. A board member asks why they aren't doing the same. An emergency strategy session gets called. Two weeks later, an "AI task force" exists. Six months after that, twenty engineers are laid off and a half-baked AI product nobody asked for is in production.

I want to talk about what's actually happening with AI adoption in the industry right now, because there are two very different stories, and most of the noise is coming from the wrong one.

---

The Bandwagon Is Real, and It's Expensive

Companies jumping on the AI bandwagon are largely doing it for one reason: optics. They're hearing stories about revolutionized workflows and massive cost savings, and instead of asking "does this apply to us?", they're asking "how do we say we're doing this too?"

Here's a pattern I've seen play out more than once. A company is paying $300 a month for a tool that works. It's not glamorous, but it does the job. Someone decides they should build their own version, with AI doing 90% of the processing. Sounds smart on a slide deck. In practice, they're now burning $1,000 a month or more in API tokens alone, before accounting for the engineering time to build and maintain it. The replacement costs three times as much and does roughly the same thing.

To offset those costs, they lay people off. To justify the decision publicly, they announce they're now an "AI-powered company." The numbers don't add up, and most of the engineers on the team know it, but the decision was made three levels above them.

---

Why This Keeps Happening

The pressure isn't coming from engineers. Most engineers I know can see through this pretty quickly. The pressure is coming from investors who want to hear the word "AI", from boards afraid of being left behind, and from leadership teams who read the same three articles everyone else did and drew the wrong conclusions.

There's also a deeper issue. Very few people at the decision-making level actually understand what AI can and cannot do. They see the demos. They don't see the failure modes. They hear "AI handles 90% of this automatically" without asking what happens in the other 10%, what the token costs look like at scale, or whether the problem they're solving even benefits from this approach at all.

When nobody in the room understands the technology well enough to push back, you end up with expensive decisions that look great on paper and fall apart in production.

---

What Good AI Adoption Actually Looks Like

The companies getting this right aren't making announcements about it. They're identifying specific problems where AI meaningfully changes the outcome, building carefully, measuring the results, and only scaling what actually works. They understand their costs. They know where their systems fail. They keep humans in the loop where it matters and automate where it genuinely makes sense.

These are also the companies hiring thoughtfully right now. Not looking for someone who has "AI" on their CV, but for engineers who understand how to build reliable systems that incorporate AI. People who know about RAG pipelines, agentic workflows, prompt reliability, cost management, and the unsexy work of making these things production-ready.

That's a different profile from what the bandwagon companies are hiring for, and it's worth understanding the difference before you accept an offer.

---

How to Tell Them Apart in an Interview

Ask specific questions. Vague answers are a red flag.

"How is your team currently using AI in production?" A company doing it properly will have a specific, concrete answer. A bandwagon company will give you a vision statement.

"What does your AI infrastructure cost to run, and how does that compare to what you replaced?" Companies with a handle on their AI usage know this number. Companies that don't are guessing.

"Where have your AI integrations failed or underperformed, and what did you learn?" Mature teams have real stories from this. Teams that haven't shipped anything real yet won't.

The answers to these three questions will tell you more about a company's engineering culture than any Glassdoor review.

---

Where This Is Heading

The companies making these mistakes are not going to admit they were wrong. What usually happens is a slow unraveling. Costs accumulate, the AI product underperforms, the engineers who actually understood the systems leave, and eventually the cuts come again. Some course-correct. Others don't recover.

The ones who get through it hire back the expertise they laid off, often at higher rates, often as contractors. This same cycle played out with cloud adoption, with microservices, and with mobile. AI is not different in this regard.

The engineers who build the most interesting things in the next decade will do it at companies that took AI seriously from the start, not the ones who announced it loudest. Those companies exist, they're hiring, and they're worth looking for.

---

Have you seen this play out at a company you've worked at? Or are you an engineer trying to push back from the inside? Drop a comment below.