Are We Still Thinking, or Just Prompting?

I use AI every day.

As a software developer, it has changed how I work in ways I would not have believed a few years ago. I explain a problem, give some context, and within seconds I have a few possible solutions in front of me.

Something breaks? Ask AI.

Need to design a new feature? Ask AI.

Not sure how to structure something? Ask AI.

There is almost always an answer.

Lately I have started wondering if that is also a problem.

Not because the answers are bad. Sometimes the problem is that the answers are very good.

We used to have to sit with problems

There was a time when getting stuck actually meant being stuck.

You read documentation. You searched Stack Overflow. You drew things on paper. You tried something, broke it, realised your approach made no sense, and started again.

It was frustrating. But something else happened during all that.

You thought.

Sometimes you would spend an hour on a problem and then realise the real problem was somewhere else entirely. Or that the library you were trying to use was not even necessary. Or that the database table you were designing did not need to exist. Or that instead of making a process faster, you could just get rid of it.

That struggle was not wasted time. A lot of the time, it was where the actual engineering happened.

Today there is a much easier option.

Prompt.

AI already knows how we build things

AI has been trained on a huge amount of human knowledge. Code, documentation, books, forum posts, tutorials, architecture patterns, open source projects, questions people asked and the answers other people gave.

That is what makes it so useful. If I ask AI how to design an authentication system, it does not need to rediscover authentication from scratch. It can pull from decades of accumulated experience.

Same with databases, APIs, caching, queues, and pretty much everything else we build.

That is real leverage.

But there is another side to it. AI is very good at showing us how problems have already been solved. What happens when we stop asking whether they should be solved that way at all?

The danger of a good answer

Say you give ten engineers the same unusual problem.

Without AI, you would probably get ten different approaches. Some would be bad. A few would work well. And one person, coming at it from a strange angle, might discover something nobody else thought of.

Now imagine all ten start the same way. They ask AI.

They will not get identical answers, but the answers will probably lean toward the same established patterns. Use this architecture. Use this database. Add a queue here. Cache this. Expose an API. Use this popular library.

These might all be perfectly good suggestions. The software gets built faster. Everyone saves time. It feels like a win.

But there is a question hiding under all of this: did we just reduce the number of possibilities we were willing to look at?

There is an idea in optimisation called a local maximum. Picture climbing a mountain in thick fog. You keep walking uphill until there is nowhere higher left to go. You look around and think, I made it to the top.

You did. You made it to the top of that mountain. What you cannot see through the fog is that a taller mountain sits a few kilometres away.

I wonder sometimes if AI makes it very easy for us to land on local maximums. We find a good solution quickly, and because it is good, we stop looking.

We might just be building better buckets

Picture a village where people walk several kilometres every morning to fetch water.

Someone decides to fix this. They design a better bucket. It is lighter, holds more water, and has a comfortable handle. Maybe AI even helps them test materials and cut manufacturing costs. It becomes the best bucket the village has ever had.

Then someone asks: why are we walking several kilometres for water every morning? And starts thinking about pipes instead.

This matters far beyond software. A lot of real innovation does not come from improving the current solution. It comes from questioning why the current solution exists at all.

We automate a form instead of asking why the form needs to exist. We make checkout faster instead of asking if checkout could disappear. We build a better dashboard instead of asking why someone needs to keep checking it. We turn a ten step process into six steps and call it innovation.

Sometimes it is. But sometimes the real breakthrough is asking why there were ten steps to begin with.

AI makes it easy to inherit assumptions

This is the part I have started noticing in my own work.

When I tell AI, "help me design the architecture for this feature," I have already made a bunch of decisions without noticing. I decided the feature should exist. I decided my description of the problem is correct. I might have already picked the technologies. I might have already pictured roughly what the solution looks like.

Then AI helps me optimise inside that box.

The better question comes before any of that: is this even the right box?

Questions I try to ask before I prompt

I am not saying stop using AI. I have no plans to. The productivity gain is too real to pretend otherwise.

But I want more space between hitting a problem and immediately asking AI to solve it. Some questions I try to ask first:

• What am I actually trying to solve here?
• Why does this problem even exist?
• What am I assuming has to be true?
• If the current solution did not exist, how would I approach this?
• Can I remove a step instead of speeding it up?
• What happens if I remove this piece completely?
• Why am I storing this information at all?
• Does this really need a database?
• Does this really need a server?
• What would make this ten times simpler?
• What would make this ten times cheaper?
• Is this even a technical problem?
• Am I solving the problem, or just improving how we currently deal with it?

You do not need to be a developer to ask these. A business owner can ask them about a company process. A designer can ask them about an interface. A teacher can ask them about how a class runs. A doctor can ask them about a workflow.

Sometimes the important question is not "how do we do this better." It is "why are we doing this at all."

AI can actually help with this too

The funny thing is, AI can be part of the answer here too. We just need to ask it different things.

Instead of "give me the best architecture for this," I can say: "Here is the problem. Do not design a solution yet. Tell me what assumptions I am making."

Instead of "how should I build this feature," I can ask: "Is there a way to get the same outcome without building this feature at all?"

Or: "Give me five completely different ways to think about this problem."

Or: "Which parts of this exist only because of some old historical or technical limitation?"

Or my favourite: "What would have to be true for us to get rid of this problem instead of solving it?"

Now AI is not replacing the thinking. It is stretching it.

And after asking those questions, there is one more thing worth doing.

Close the AI. And think.

Answers are getting cheap

This might be the biggest shift AI is bringing.

For most of human history, getting answers was hard. Knowledge was scarce. You needed the right books, the right teachers, the right experts, or years of experience.

The internet already changed that a lot. AI has pushed it even further. We are heading into a world where answers are almost free. You can ask a hard technical question and get a solid response before your coffee gets cold.

That is genuinely amazing. But when something becomes abundant, something else usually becomes scarce.

Maybe the scarce thing now is not answers. Maybe it is good questions. Maybe it is the patience to actually sit with a problem. Maybe it is knowing a bad assumption when you see one. Maybe it is the willingness to skip the obvious answer and keep looking anyway.

Having a calculator lets me calculate faster. It does not make me better at maths. Having GPS gets me somewhere faster. It does not improve my sense of direction. And having AI that gives me great answers does not automatically make me a better thinker.

That part is still on us.

AI can show us how thousands of people solved a problem before. That is one of its biggest strengths. But maybe one of the most important skills in the AI age is knowing when those answers are not enough.

Somewhere out there, someone is building a better bucket.

Hopefully someone else is still asking why we need the bucket at all.