AI Made It Easy to Build. That Became My Problem.

Almost every problem now looks solvable

I naturally notice problems and start thinking about how they could be solved.

A messy delivery process. A repetitive expense workflow. A server I wish I could control differently. A better way to package, test, or distribute software.

Earlier, most of these thoughts would have remained thoughts.

Now they can become products.

Prompting is only a small part of it. Over time, I have become better at what I think of as loop engineering and graph engineering.

Instead of asking AI for one answer, I think about the whole system.

Which agent should do what? What context does it need? How should it verify its own work? When should another model review the output? What happens if something fails? Which tools need to be connected? How does information move from one step to the next?

Once you learn how to design those loops, handoffs, checks, and workflows, AI becomes much more than a chatbot.

It becomes a way to turn a thought into a functioning system.

That is incredibly powerful.

It is also where my problem begins.

Every solvable problem starts looking like a business

When you can build quickly, it becomes very easy to convince yourself that every useful idea should become a product.

I do this constantly.

I see a problem. I imagine the solution. I’ve done this before with things like Muster and Vaqelio, jumping straight to the website, pricing, users, integrations, and the larger platform it could eventually become.

Before I know it, I am thinking about another business.

The old question was:

“Can I build this?”

The new question should be:

“Should I build this?”

AI is very good at helping me answer the first question.

It cannot answer the second one for me.

That requires judgment, focus, conversations with real people, and the willingness to stay with one problem after the initial excitement is gone.

Unfortunately, a new idea is always more exciting than marketing the current one.

Building is the comfortable part

Building gives me an immediate feedback loop.

I ask for something. The screen changes. A feature starts working. A test passes. The product gets deployed.

It feels like progress because I can see the result.

Marketing is completely different.

Marketing means repeating the same message several times.

It means posting when nobody may respond.

It means reaching out to people, asking questions, hearing no, discovering that my explanation is unclear, and accepting that something I built may not matter as much to other people as it does to me.

There is no instant reward.

When another idea appears, building it feels productive. Sometimes it is. Sometimes it is just a sophisticated way of avoiding the uncomfortable work that the current product needs.

I can tell myself that I am still working.

I can research another market, design another architecture, create another landing page, or build another prototype.

The activity is real. The output is real.

But that does not necessarily mean I am moving forward.

Overthinking makes this even worse. With AI, I can now research every possible competitor, compare every technical approach, rewrite the positioning endlessly, and explore five different versions of the same product.

AI can help me make a decision.

It can also help me postpone one indefinitely.

Speed is not the same as value

A distinction from a recent METR study has stayed with me.

In a 2026 survey of 349 technical workers, respondents estimated that AI had increased the value of their work by roughly 1.4 to 2 times. Their estimated speed increase was even higher, around 3 times.

But the researchers were careful about the difference between speed and value. AI can make certain activities dramatically cheaper, which can lead people to do more of those activities even when they are not the most valuable things to do. The researchers also warned that self-reported productivity can be unreliable. The study is worth reading.

That describes my experience almost perfectly.

AI allows me to create more.

It does not automatically make everything I create valuable.

A working prototype built in two days can still be a distraction.

A polished website can still have no users.

Ten products that are 80 percent complete may create less value than one simple product that has been placed in front of the right ten people.

AI has not removed the bottleneck.

It has moved it.

The bottleneck used to be production. Now it is judgment, focus, distribution, and deciding what deserves my attention.

Has AI made me dumber?

Sometimes, honestly, it feels like it has.

I occasionally ask AI to phrase something before I have properly formed my own thought.

I ask it to summarize information before I have read the original.

I can ask it to debug something before I have tried to understand why it broke.

If I do that too often, I am outsourcing the struggle that would normally force me to think.

A study from Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers and collected 936 examples of people using generative AI at work. It found that higher confidence in AI was associated with less critical thinking. It also found that the nature of thinking changed. People moved away from producing everything themselves and spent more effort verifying information, integrating responses, and supervising the task. That research reflects the tension quite well.

I do not think this proves that AI makes people stupid.

But it does show that using AI well requires a different kind of intelligence.

I may not need to write every line of code myself, but I need to understand what the system is supposed to do.

I may not need to perform every piece of research manually, but I still need to judge whether the conclusion makes sense.

I may not create every individual output, but I am responsible for the outcome.

On my best days, AI does not replace my thinking. It increases the size of the problems I can think about.

I can explore architectures, test assumptions, compare approaches, orchestrate several agents, and build systems that would previously have required far more time or a larger team.

On my worst days, it gives me a very convincing way to avoid thinking deeply.

So I do not believe AI has made me uniformly smarter or dumber.

It has made it easier to skip thinking, and it has made it possible to attempt much harder things.

Which one wins depends on how I use it.

The skill I need now is restraint

For a long time, I believed the valuable skill was learning how to build with AI.

It still is.

But the scarcer skill may be knowing what not to build.

I need to become comfortable putting a new idea into a backlog instead of immediately turning it into a repository.

I need to treat distribution as part of building, not something that begins after the product is “finished.”

I need to speak to people before adding another feature.

I need to ask whether anyone is using what already exists, whether the problem is painful enough, and whether somebody would actually pay to solve it.

I also need to use AI for the work I tend to avoid.

It can help me turn product decisions into useful content. It can help me prepare outreach, analyze customer conversations, improve demonstrations, and explain what I am building more clearly.

If I use AI only to build, I will simply become faster at creating things nobody knows about.

This blog is part of the correction

The irony is that writing this is one of the things I would usually postpone.

I would rather improve a product, test another agent, or explore the next use case.

Brand building and distribution do not come as naturally to me as building does.

That is exactly why I need to do them.

This website and these blogs are my attempt to document what I am working on, what I am learning, where I am wrong, and how my thinking changes over time.

Publishing forces me to stop building for a moment and explain why any of it matters.

Because if I only keep building, I will end up with an impressive collection of prototypes instead of something people genuinely use.

The next breakthrough for me probably will not come from a better prompt, a more capable model, or another agent.

It will come from staying focused on one problem long enough for the work to reach people.

AI has made building cheaper.

It has not made attention cheaper.

It has given me more leverage than I have ever had before. Now I need to get better at aiming it.

I can build almost anything.

I cannot build everything.

That may be the most important thing AI has taught me so far.