AI Is Not the Bubble. Our Expectations Might Be. 🤔

Let's be honest.

AI has gone a little crazy lately.

You open LinkedIn. Someone has built an entire startup using AI.

You open YouTube. Someone claims software engineers will be unemployed by 2027.

You open X. Someone is arguing that the latest AI model is either going to change humanity forever or is just glorified autocomplete.

And then you open GitHub...

Another 500 AI-powered wrappers. 😭

Jokes aside, something genuinely interesting is happening.

AI is getting better. Companies are investing billions. Developers are experimenting with coding agents, and even people who have never written a line of code are building applications.

But there's one thing I've been wondering:

Are we actually witnessing an AI revolution, or are we slowly building expectations that reality won't be able to keep up with?

I don't think AI is going anywhere.

But I do think we need to have a conversation about the gap between what AI promises and what it actually delivers.

Let's dig into it.

1. First things first... AI IS REALLY GOOD. 🚀

Before calling everything hype, let's give credit where it's due.

AI has made some ridiculous progress.

Things that seemed almost impossible a few years ago are now accessible through a simple chat interface.

Need help debugging code? AI.

Need to understand a complicated research paper? AI.

Want to build a prototype over the weekend? AI.

Want to generate an image of a penguin running a software company?

Well... AI can probably do that too. 🐧

And the adoption numbers are equally impressive.

According to Stanford University's 2026 AI Index Report:

• 88% of surveyed organizations reported using AI in 2025.
• Generative AI reached approximately 53% population adoption within three years.
• US private AI investment reached approximately $285.9 billion in 2025.

Those aren't small numbers.

AI has moved far beyond being something only researchers and tech enthusiasts experiment with.

But here's something I found particularly interesting.

AI can perform exceptionally well at one task and struggle with something that looks ridiculously simple.

Researchers call this the jagged frontier of AI.

For example, a model might solve a complicated programming problem but still struggle with an unfamiliar interface or a seemingly simple visual task.

That's what makes AI interesting.

It's not that AI is bad. It's that its capabilities aren't evenly distributed.

And I think this is something we often forget when watching those impressive 30-second demos.

A demo shows you what a model CAN do.

It doesn't necessarily tell you how reliably it can do it 100 times.

Source: Stanford AI Index 2026

2. Everyone is becoming productive... so where's the money? 💸

Now this is where things get interesting.

One of the biggest arguments in favour of AI is productivity.

Developers write code faster.

Marketing teams generate content faster.

Analysts process information faster.

Basically, everyone is getting a productivity upgrade.

But there's an interesting contradiction.

McKinsey's State of AI 2026 reported:

• 80% of respondents experienced improvements in individual productivity.
• Only 37% reported at least some enterprise-level financial impact from AI.
• Just 6% qualified as AI high performers, with significant financial impact from AI.

Wait... what?

If everyone is becoming more productive, why aren't companies seeing proportionate financial benefits?

Here's how I think about it.

Imagine you have a development team of five people.

Before AI, building a particular feature takes two weeks.

Now, with AI assistance, the initial implementation takes three days.

Amazing! 🎉

But then comes the fun part.

Someone needs to review the generated code.

Someone needs to fix the unexpected bugs.

Someone needs to test the edge cases.

Someone needs to make sure the implementation doesn't accidentally expose customer data.

And someone still needs to maintain the whole thing six months later.

Suddenly, the original three days look a little different.

I'm not saying AI doesn't save time. It absolutely can.

But:

Reducing the time required to produce something doesn't automatically reduce the time required to make it work properly.

And that's a distinction businesses are still figuring out.

3. Developers, are we actually getting faster? 👨‍💻

This is probably my favourite part of the whole AI discussion.

Remember when building a full-stack application was considered a decent engineering achievement?

You needed to understand frontend development, backend architecture, databases, authentication, APIs, deployment...

Now you can open an AI coding agent, describe what you want, and watch it start modifying your entire codebase.

Sometimes it works beautifully.

Sometimes you ask it to fix a button and it decides to rewrite your entire authentication system.

Classic. 💀

But seriously, coding agents are becoming increasingly capable.

So I came across a rather interesting experiment.

In July 2025, METR published a randomized study involving 16 experienced open-source developers working on 246 real-world tasks.

The result?

Developers using the early-2025 AI tools tested actually took 19% longer to complete their tasks.

Even more interesting:

Before using the tools, developers expected them to make them 24% faster.

After the experiment, they still believed AI had made them faster.

That's quite a difference between perception and measurement.

Now, there's an important update.

In February 2026, METR explained that its follow-up experiment suffered from selection bias and couldn't reliably estimate the productivity gains from newer AI tools.

So no, this doesn't prove that modern AI coding agents make developers slower.

But it does make us think about something.

Are we measuring productivity by how much code we generate, or by how much useful software we actually deliver?

Because those are two very different metrics.

Sources: METR's original experiment · February 2026 update

4. Okay, but why does everyone keep calling it a bubble? 🎈

Let's move away from coding for a moment.

Look at the amount of money entering the AI ecosystem.

According to Stanford's 2026 AI Index, US private AI investment reached approximately $285.9 billion in 2025.

And the investments aren't limited to startups building chatbots.

We're talking about:

• Massive data centres.
• GPUs and computing infrastructure.
• Electricity and cooling requirements.
• Training increasingly powerful models.
• Building the infrastructure to serve millions of users.

Everyone wants to be part of the AI race.

And honestly, who wants to be the company that missed the next internet?

But here's the question:

Does spending billions on a technology guarantee that every company investing in it will make money?

Of course not.

And this brings us to an interesting comparison.

Remember the dot-com bubble? 🌐

Image suggestion: Historical NASDAQ Composite chart, 1994–2005.

During the late 1990s, the internet was transforming the world.

Companies were moving online. Startups were attracting massive investment. Everyone wanted to be part of the internet revolution.

And then came the crash.

A lot of companies disappeared.

But here's the interesting part.

The internet didn't disappear with them.

The technology continued to develop. New businesses emerged, and the internet eventually became an essential part of everyday life.

So what does this tell us about AI?

Maybe AI can become one of the most important technologies of our generation while many AI companies still fail.

Maybe some valuations are unreasonable.

Maybe some companies are spending more than they can justify.

Maybe a few products that are incredibly popular today won't even exist five years from now.

And none of that necessarily means AI itself is a failure.

A technological revolution and an investment bubble can exist at the same time.

That's the distinction I think gets lost in a lot of these discussions.

5. The real problem? We're confusing potential with reality. 🤯

Let's look at some of the expectations circulating around AI.

And there's one thing I'd like to emphasize.

Not every problem needs AI.

Imagine building an application that uses an LLM to calculate someone's monthly expenses.

Why?

A simple calculation would probably be cheaper, faster, and more reliable.

Yet we sometimes see people adding AI to everything simply because they can.

It's like putting a Ferrari engine inside a bicycle.

Cool? Absolutely.

Necessary? Probably not. 😂

The best engineering solution isn't always the most complicated one.

Sometimes boring technology is exactly what you need.

6. So... should developers be worried? 👀

I think this is where the discussion becomes personal.

There are developers who are worried about losing their jobs.

There are students wondering whether learning programming is even worth it anymore.

And there are people who believe learning how to use AI is now more important than learning how to code.

I don't completely agree with either extreme.

Personally, I think developers should learn to use AI effectively.

Ignoring it would be foolish.

But depending on it for everything is equally dangerous.

Think about it.

If AI generates your database schema, do you understand normalization?

If AI writes your API, do you understand authentication and authorization?

If AI creates your application architecture, can you explain why those particular design decisions were made?

If the answer to all of these is no, then you might have a working application, but you probably don't have a strong understanding of what you've built.

And that's a problem.

Because when something breaks, you can't simply ask your AI agent to fix it 15 times and hope for the best.

Well... you can, but good luck in production. 💀

I think the value of engineering fundamentals will remain extremely high.

In fact, as code generation becomes easier, knowing how to evaluate, validate and maintain that code becomes increasingly important.

7. My point? AI might be real. The hype might be real too.

I don't want AI to fail.

I genuinely enjoy seeing what people are building with it.

The pace of progress is fascinating, and I think we're still relatively early in understanding what this technology can become.

But I also don't want to blindly accept every prediction made about it.

Some companies will build incredible products.

Some will burn through millions without finding a sustainable business model.

Some developers will become significantly more productive.

Others will spend more time reviewing generated code than writing it themselves.

Some AI agents will become dependable tools.

Others will remain impressive demos that look great on LinkedIn. 😂

And that's okay.

Not every experiment needs to become a success.

What bothers me is when we start treating the future as if it has already happened.

The model is the demo. The product is everything that happens after the demo.

That's where I think the real competition will be.

Not just in building smarter models, but in making them useful, reliable, affordable, and valuable in the real world.

And perhaps the biggest skill in the AI era won't be knowing every new tool.

It might be knowing which tools are actually worth using.

This is the first article in The AI Reality Check — a series exploring AI's promises, limitations, economics, and its impact on software engineering.