AI can get you 90% of the way to a spec in seconds

AI can get you 90% of the way to a spec in seconds.
That might be the problem.

There’s something I’ve been thinking about when it comes to AI-generated specs.

Yes, tools like spec kits can get you 80–90% of the way from a blank page to something structured. That’s useful. Starting from something is psychologically easier than starting from nothing. Structure feels like momentum. It feels like progress.

But there’s a subtle risk most people don’t talk about.

When AI generates the first version of your spec, architecture, or plan, it doesn’t just give you content. It gives you a frame. And that frame quietly shapes everything that comes after.

Even if you tell yourself “this is just a draft,” it becomes the anchor. The first model. The first interpretation of the problem. From that moment on, you’re reacting to a structure instead of originating one. You’re editing instead of inventing.

That shift matters.

There’s a well-documented cognitive bias called anchoring: the first number or structure you see influences your later decisions, even if you know it’s arbitrary. AI-generated plans work the same way.

The AI suggests a three-tier architecture. Even if you reject it, you’re now thinking in tiers. The frame is set.

The initial flow, the proposed data model, the suggested constraints—they narrow the solution space before you’ve explored it yourself. And the more polished and confident the output, the stronger the gravity.

The comfort is real. Going from blank canvas to “90% there” feels productive. But that comfort can also be false certainty.

When you over-spec early—especially with AI—you over-invest in a particular logic and direction. You become subtly committed to a path you didn’t fully author.

There’s no such thing as “temporary” exposure in this context. Once a base is generated, it imprints. Even if you reject parts of it, your mental model has already been influenced.

The danger isn’t that AI is wrong. The danger is that it frames the problem before you’ve wrestled with it yourself.

And framing is the highest-leverage part of design.

Think about what you’re actually outsourcing.

For execution-heavy tasks—implementing a known pattern, scaffolding boilerplate, formatting documentation—AI-generated specs can be incredibly efficient.

But for product shaping, architecture decisions, and system design? The first move matters. The first 10% defines the shape of everything that follows.

If that move is outsourced, you’re no longer the originator. You’re the editor.

You’re working within a solution space the AI defined. You’re optimizing variables the AI chose. You’re solving a problem the AI framed.

That’s not necessarily bad. But it’s different.

And most people don’t realize the trade they’ve made until they’re deep into implementation and something feels… off. The structure doesn’t quite fit. The abstractions feel borrowed. The system lacks coherence.

Because coherence comes from wrestling with the problem yourself. From discovering the constraints. From asking what the material wants to be—not from inheriting someone else’s (or something else’s) interpretation.

The efficiency paradox: You saved time on the spec. But you outsourced the highest-value part: defining what problem you’re actually solving and what shape the solution should take.

You can move fast on execution. But you’re executing someone else’s plan.

So maybe the real question isn’t whether AI can get you 90% there.

It’s whether you want your first 10%—the part that defines the shape of the system—to come from you or from the model.

Because that first 10% is where all the leverage is.

That’s where you decide:

• What abstraction makes sense
• What constraints are actually binding
• What structure fits your specific context
• What the system wants to become

You can be efficient or you can be original. With AI-generated specs, you rarely get both.

I’m not saying never use AI to generate specs. I’m saying be conscious of what you’re trading.

If you’re executing a well-understood pattern in a known domain, generate away. The frame doesn’t matter because the frame is already established.

But if you’re shaping something new, solving an unfamiliar problem, or designing system architecture—consider whether that first frame should come from you.

Because once it’s generated, you can’t unsee it. The anchoring has already happened. The solution space has already narrowed.

The alternative isn’t starting from nothing.

The alternative is starting from your own exploration. Sketching the constraints yourself. Asking the questions yourself. Discovering what the problem actually is before defining what the solution should be.

That’s slower. But it’s the only way to build something that’s actually yours—structurally, conceptually, architecturally.

The spec can come later. After you’ve wrestled with the material. After you understand what it wants to be.

But if the spec comes first, you’ve already given away the part that matters most.