Is Gen AI Approaching Bubble Territory?

Gen AI looks like a bubble, but people are still using it to get real work done.

It’s the internet, cloud, and crypto rolled into one. Everyone’s betting big. But when you peel back the hype, a harder question emerges: is this transformation, or just the most expensive game of follow-the-leader in tech history?

The Emperor’s New Algorithm

We’re living through the most overhyped, under-delivered technology moment since the dotcom boom. But here’s the kicker: unlike 1999, this bubble might actually have substance underneath all the noise.

Every tech CEO is suddenly an AI evangelist. Every startup pitch deck has “AI-powered” slapped on slide two. Every engineering manager is asking their team why they’re not “leveraging GenAI for 10x productivity gains.”

It’s exhausting. And it’s mostly wrong.

The Numbers Don’t Add Up

Microsoft, Google, and Meta claim 30% of their production code comes from AI. The useful question is what kind of code counts.

What kind of code are we talking about? Boilerplate? Test scaffolding? The mundane glue that holds systems together? Or is it changing product behavior that users can actually feel?

My bet: it’s the former. AI excels at the repetitive, the predictable, the already-solved. That’s valuable, but it’s not revolutionary. It’s evolutionary automation.

The capital cycle around model companies, chip vendors, cloud providers, and investors is hard to read cleanly. Some of the revenue is real usage. Some of it looks like money moving around the same ecosystem to justify the next round of spend. Ed Zitron’s Better Offline Monologue nails it, pointing out how the math is comically broken.

Mandates Are Not Adoption

Brian Armstrong’s Cursor and Copilot mandate, documented by CIO) is a useful example of confusion. A tool mandate can create compliance, but it does not prove the tool changed the work.

This is the same playbook every bubble uses: create urgency through fear. “Adapt or die.” “The future is now.” “Your job depends on it.”

Reality check: companies still need competent engineers, with or without AI copilots.

ATMs were supposed to obliterate teller jobs, but the number actually increased as banks expanded their footprint. Automation usually changes the job before it removes the job.

The 70% Who Aren’t Panicking

Stack Overflow’s data reveals something interesting: 70% of developers don’t see AI as a career threat.

That is not denial. It is pattern recognition from people who have watched tool cycles come and go.

They’ve seen this movie before: with NoSQL, with microservices, with blockchain. New tools emerge, some stick, most fade. The fundamentals remain: solving real problems for real users with maintainable code.

Where the Real Battle Lines Are Drawn

The AI divide isn’t between “AI adopters” and “AI skeptics.” It’s between companies that can afford to experiment and those that can’t.

Big corporations have the resources to retrain their workforce. They can absorb the learning curve, the false starts, the inevitable productivity dips that come with new tools.

Small businesses? They need digital natives who already speak AI. Big corporations can retrain; small ones can’t. Their edge comes from hiring people fluent in AI from day one. For them, AI fluency is cheaper to hire than to teach.

The Useful Part Is Access

The useful part is not 10x individual productivity. It is reducing the cost of tasks that used to require specialized help.

Remember research before Wikipedia? You walked to a library. You had encyclopedia collections. Knowledge had friction.

AI tools are removing friction from tasks that used to require specialized knowledge. That’s powerful, but it’s not magic. It’s infrastructure.

Bubble? Yes. Crash? Maybe Not.

This feels like a bubble because it is one. The investment levels are unsustainable. The promises are overblown. The timeline expectations are divorced from reality.

But unlike the dotcom crash, the underlying technology isn’t vaporware. Large language models work. They solve real problems. The use cases are expanding, not contracting.

The crash, when it comes, won’t kill AI. It’ll kill the companies that confused venture capital with product-market fit. It’ll humble the CEOs who promised the moon. It will punish companies whose AI story was stronger than their product.

What to Check

Before you go all-in on the AI revolution:

1. Question the metrics. When a company claims “AI-generated code,” ask what kind of code and what kind of problems it’s solving.
2. Follow the money. Who’s actually making money from AI? Really, there are two categories: the picks-and-shovel providers like NVIDIA and Broadcom, and the big tech incumbents (Microsoft, Google, Amazon, Meta) using AI to boost existing profits.
3. Listen to the users. Developers, designers, writers, the people actually using these tools daily, have the most honest perspective.
4. Remember the fundamentals. Good software is still about solving user problems, not showcasing cool technology.

The Real Disruption Is Invisible

The companies winning with AI aren’t the ones shouting about it. They’re quietly integrating it into workflows, removing friction, solving boring problems faster. Take Notion, which folded AI into documents without fanfare but made millions of users stickier. Or Workflowy, which has been shipping like crazy lately.

They’re treating AI like any other tool in the toolkit: useful when appropriate, ignored when not. They’re not trying to replace humans; they’re trying to amplify human capabilities.

This pragmatic approach won’t generate TechCrunch headlines. It won’t land keynote speaking slots. But it will make normal work cheaper, faster, or less annoying. That is where adoption sticks.

What This Means for Engineering Leaders

Stop chasing the AI hype cycle. Start identifying where your team spends time on solved problems.

Code generation, documentation, and test creation are good AI candidates because they are predictable, not because they are exciting.

Invest in education, not tools. The specific AI platforms will change. The ability to evaluate and integrate new capabilities won’t.

Build systems that can incorporate AI outputs without depending on them. Treat generated code like boilerplate: a useful starting point that’s often solid, but still requires review and context.

The Bottom Line

The bubble is real. So is the work underneath it. Unlike crypto or the metaverse, AI is actually solving real problems for real users right now.

The crash will come. The hype will fade. The tourist money will move to the next shiny object.

What will remain: better tools, clearer use cases, and engineering leaders who learned to separate signal from noise.

AI does not need the bubble to be useful. It needs the bubble to stop being the only story.