DEV Community

michielinksee
michielinksee

Posted on

I let AI build my products for months — then I couldn't tell what any of them were anymore

#ai #showdev #programming #productivity

Claude Code and Cursor make you ship fast.

But once I was running several products in parallel — and across several different LLMs (Claude Code, Cursor, ChatGPT) — a different problem showed up:

  • I'd forget what I'd decided (the human)
  • A design I locked in via Claude Code, the Cursor session didn't know about (the LLM)
  • The README and the code drifted apart
  • The CLI changed, but npm and the docs stayed old
  • Neither I nor the AI could say which part of the product we were even touching

Every single change was correct. But the product as a whole kept sliding away from what it was supposed to be.

That's rough.

Building just one product? CLAUDE.md is probably enough. This is for people juggling several.

There's a name for this now: intent drift — the product slowly diverging from what you meant to build.

The exact moment it bites

You ask Claude Code:

fix the CLI setup flow

It does. Perfectly.

But to actually match that change, you probably also needed to touch:

  • README
  • the npm description
  • the GitHub install steps
  • docs
  • the landing page
  • the demo screenshot

The AI fixes the spot you asked about. That's it.

So you get:

The engine is fine. The install steps are stale. Users get stuck.

That's scarier than a bug. The code runs — the story around it is wrong.

Here's the tool catching exactly that — the README promises a --export flag the code doesn't have, and it shows the blast radius too:

Linksee catching that the README's --export flag is missing from the code, with the blast radius

A product has a map

A product grows along a path:

discovered → understood → tried → installed → kept → paid → expanded
Enter fullscreen mode Exit fullscreen mode

And under each step sits a stack of surfaces:

LP / README / npm / GitHub / CLI / MCP / dashboard / docs / Stripe / support

While coding, you want to know: "where on this map am I, and if I change this, what else moves?" But running several products across several LLMs, that map falls out of your head — and nobody is tracking it. That's the drift.

So I'm building Linksee Memory

A local-first tool that records the decisions, the implementation, and the drift as you build with AI — so you can see them later.

It started as a memory tool. Now I think of it differently:

A shared map of the project — for the human and the multiple AIs.

Three commands.

1. What's drifting? 

npx linksee-memory map status
Enter fullscreen mode Exit fullscreen mode

Lists what's out of sync:

  • README says it, code doesn't
  • CLI changed, docs are old
  • LP and README disagree
  • a flow you planned, then abandoned

2. What do I fix next? 

npx linksee-memory map next
Enter fullscreen mode Exit fullscreen mode

3. Show me the whole map 

npx linksee-memory map blueprint
Enter fullscreen mode Exit fullscreen mode

Bonus: npx linksee-memory map where README.md tells you where a file sits, and what changing it would touch.

It all lives in a local SQLite file — so Claude Code, Cursor, ChatGPT, Codex and Gemini share the same map. No cloud, no API key, MIT.

Why it helps

The scariest thing in AI development isn't the code breaking.

It's the code working while the product's story — install, docs, promises — quietly drifts.

Try it 

npx linksee-memory init
Enter fullscreen mode Exit fullscreen mode

Everything stays local. Nothing leaves your machine.

A GitHub star genuinely makes my day 🙏

I'd love your take

If you run multiple products / multiple LLMs — have you hit the "wait, what was this even for?" moment?

  • Was it a memory problem, or a map problem?
  • Of "what's drifting," "what's next," and "the whole map" — which would you actually use?

Drop a comment, or find me on X (@michielinksee).

Solo founder in Singapore, building in public.

Top comments (0)