We can build software faster than ever before.

You can describe an idea to an AI agent, give it a few instructions, and watch something that might have taken weeks or months begin taking shape in minutes.

You'd expect an explosion of creativity: more unusual ideas, ambitious experiments, and people attempting things that previously seemed impossible.

Yet sometimes I feel we're seeing the opposite.

Another AI wrapper. Another dashboard. Another application that looks remarkably similar to something that already exists.

Of course, people are doing genuinely interesting work. I'm not suggesting innovation has stopped. Maybe I'm just looking in the wrong places.

But I keep wondering:

If we now have some of the most powerful tools ever created, where is the spark?

Building isn't the same as imagining

There's a difference between being able to build something and imagining something worth building.

Generative AI is reducing the effort required to turn ideas into working systems. You don't necessarily need to understand every implementation detail before experimenting with an idea. That's incredible.

But making something easier to build doesn't automatically make us more imaginative. If we keep asking these systems to reproduce what already exists, why should we expect different outcomes?

I think we risk confusing productivity with creativity. Producing ten applications doesn't necessarily mean you've explored ten original ideas.

And perhaps that's where our thinking needs to change.

For a long time, much of our attention has gone into figuring out how to build things. As machines become increasingly capable of handling implementation, what happens to all that attention?

Do we simply produce more of the same things? Or do we start:

  • questioning assumptions,
  • connecting unrelated ideas, and
  • investigating problems we've accepted as normal?

What's stopping someone from vibe coding Google?

Here's something I've been thinking about.

What's stopping someone from opening Claude Code, Codex, or another coding agent and asking it to build Google? They'd need:

  • a web crawler,
  • a search index,
  • a ranking algorithm,
  • a query interface, and
  • perhaps even an advertising system.

Nothing is stopping them from trying, and they could probably build a functional search engine prototype surprisingly quickly.

But have they built Google? Not quite.

Google isn't simply a collection of source files. Behind its search engine are enormous infrastructure investments, continuously updated indexes, ranking systems, years of operational knowledge, and an entire ecosystem.

Writing software that resembles Google is one problem. Building something capable of competing with it at scale is another.

Are software moats really disappearing?

This is why I find the argument that software moats are disappearing somewhat incomplete.

Yes, writing code is becoming cheaper. Some products will become much easier to replicate, and some existing advantages will weaken.

But the value of a system has never been entirely about how difficult its code is to write. These still matter:

  • reliability,
  • architecture,
  • distribution,
  • accumulated knowledge, and
  • the ability to operate under real-world conditions.

None of these advantages is permanent. AI might eventually make many of them easier to reproduce, too.

But making something easier to build doesn't necessarily make it easy to replace.

The obsession to keep pushing

There's something else I think matters enormously: obsession.

Not necessarily spending endless hours writing code yourself, but having an almost unreasonable desire to understand a problem and keep pushing a system beyond its limitations.

Imagine two people with access to exactly the same AI tools.

The first asks an agent to build an application. It works, so they launch it and move on.

The second keeps questioning everything:

  • Why does it work this way?
  • Why does this limitation exist?
  • What happens if we change the architecture?
  • Could we approach the entire problem differently?

Both have the same tools, but they might produce radically different outcomes.

You might no longer need to implement every component manually, but you can still be the person directing the experiments, questioning the results, and deciding what to try next. And I find that fascinating.

Instead of spending weeks implementing a single approach, you might be able to investigate several alternatives, discover why they fail, and keep refining your understanding.

Of course, obsession alone isn't enough. You can spend years perfecting something nobody needs. But combine it with curiosity, technical understanding, judgment, and a willingness to experiment, and there's still something valuable there.

The first working implementation shouldn't necessarily be the end of your imagination. Sometimes it's where the interesting questions begin.

This goes beyond software

I don't think this opportunity is limited to software. Think about hardware, robotics, energy, manufacturing, or scientific research.

Imagine designing a robotic hand. You describe what you want, use AI to explore mechanical designs, simulate their movements, and generate components for a prototype. Eventually, you build one.

It works, but it struggles to pick up a tiny screw.

What happens next?

Do you accept the limitation, or do you start questioning the finger geometry, sensors, materials, and control system? Perhaps an entirely different mechanism would work better.

AI could help you investigate those possibilities, but someone still has to recognise the problem, test the ideas, and understand what the results mean.

And unlike software, the physical world has constraints that don't disappear because an AI generated a convincing design. Materials cost money. Components wear out. Manufacturing takes time. Physics doesn't care how impressive your prompt is.

Still, as these tools improve, individuals and smaller teams may gain the ability to attempt increasingly complex projects. There are machines we haven't invented, systems we haven't connected, and problems we haven't investigated deeply enough.

Not everything has to become another application.

A forced evolution in how we think?

I sometimes wonder whether we're approaching a point where a kind of forced evolution in our thinking becomes necessary. Not biological evolution, but a change in how we approach problems.

We've spent years learning how to use computers to accomplish tasks. Now computers are beginning to handle increasingly complex tasks on our behalf.

Perhaps this means we need to become more deliberate about exercising our creativity and logical reasoning: to ask better questions, challenge assumptions, and understand systems rather than merely operate them. To see possibilities where others see limitations.

I'm not suggesting that implementation no longer matters, or that technical understanding becomes irrelevant. Quite the opposite.

The better you understand something, the more effectively you can question it. And the more capable our tools become, the further those questions might take us.

But there's another possibility worth considering.

What if increasingly capable AI makes it easier for us to avoid thinking altogether?

We could become so comfortable accepting generated solutions that we stop investigating why they work, whether they're correct, or whether a better approach exists.

I don't think that's inevitable. But it's a choice we'll increasingly have to make.

Where is the spark?

There is still so much left to solve: healthcare, energy, robotics, biology, computing, and countless problems we probably haven't even recognised yet.

I don't think we're running out of ideas. I think we might be underestimating how much remains possible.

And perhaps the opportunity presented by generative AI isn't simply to build the same things faster. It's to explore things we previously couldn't afford to attempt.

We've made it easier than ever to turn an idea into something real. Maybe it's time we became more ambitious about the ideas themselves.

The machines are getting better at building. Are we getting better at imagining?