Why I went the opposite direction from spec-first

Everyone’s rushing to AI tools that promise “comprehensive specs in one prompt.”

I went the opposite direction.

Not because spec-first is wrong - but because it’s not how I think.

Let me be clear upfront: there’s nothing wrong with spec-first tools. If that workflow matches how you build, use it. This isn’t a critique of anyone’s process.

But I tried. I really tried to embrace the spec-first approach. The more I used it, the more I felt my creativity stifled. I can’t work fixed, rigid, and funneled. It’s just not me.

Here’s what I’ve built instead: huge codebases with no noticeable context loss. The AI talks back like it is the project’s mind and brain because I’ve positioned it that way. And that’s allowed me to develop software like a designer.

It’s why I’ve come to prefer this over traditional tools like Figma. I’m not prototyping in a separate tool and translating to code. I’m designing directly in the material, working code, with AI maintaining the continuity that lets me stay in flow.

My actual process looks like this: scratchpad at hand, flow charts mapped out, understanding constraints, iterating back and forth. And I don’t give two shits about context windows because they’re never an issue.

Out of context? The AI auto-checks the index and goes right back at it like nothing went missing.

Context windows are great, but that’s the beauty of SR-SI: as long as you have a reasonable window and a capable model, you don’t really care about the limits. The system handles recovery automatically.

This workflow matches how I think as an architect. Sketch, test, refine, repeat. SR-SI removes the friction from that loop by handling the memory burden so I can focus on creative iteration.

The real unlock isn’t better specs. It’s no translation layer between design thinking and implementation. I think in systems and flows and relationships. The AI maintains those relationships. I keep designing. The code emerges from that process, not from executing a predetermined plan.

So when should you care about this approach?

If you’re someone who thinks in iteration, who designs by building, who needs to stay in creative flow across weeks or months - this might resonate with you.

If spec-first feels constraining to you the way it did to me, there’s another path.

But if spec-first works for your brain, keep using it.
Different minds need different tools.

The question isn’t “which approach is better.”
It’s “which approach matches how you actually think?”

For me, I need the AI to remember the structure we’re building together. To know why we made that tradeoff, where that constraint came from, how these components interlock. I need dialogue with the material that doesn’t come with 50 first dates syndrome.

To tie it back to Louis Kahn when I first kicked off this series: I need to ask the brick what it wants to be - and I need the brick to remember our conversation.

That’s what SR-SI gives me: an architect’s process.

SR-SI: The methodology that gives AI persistent memory across any long-running project