Between what you mean
and what gets built,
something gets lost.
Not because AI isn't capable.
Because intent was never structured before building started.
Natural language is ambiguous. The same words mean different things to different readers. When you describe what you want to build, AI picks an interpretation and acts on it.
If that interpretation drifts from yours, the drift compounds. Every session adds distance between what you originally meant and what is being built. By the time you notice, unwinding it is harder than starting over.
Most tools address this after the fact — better prompts, auto-memory, clearer instructions mid-build. These help. They don't fix the root: your intent was never structured precisely before building started.
Ada is the step before.
You tell Ada what you want to build. Ada asks the questions
that surface what you actually mean — not technical questions,
questions about what the thing should do and be.
Then Ada produces the files Claude Code reads at the start
of every session: CLAUDE.md, agents, hooks.
Claude Code's auto-memory records what it observed as it worked. Ada records what you intend before it starts. One is retrospective. One is prospective. They are not the same thing.
The result: less distance between what you described and what gets built. Not zero distance. Less.
- 01
You describe what you want to build.
In your language. At whatever level you think in. Ada does not ask about frameworks, libraries, or implementation.
- 02
Ada surfaces what's ambiguous and asks about it.
What does this need to do? What does it not? Who uses it and how? The minimum set of questions needed to resolve what's unclear.
- 03
Ada produces the structured context.
CLAUDE.md, agents, hooks — files on disk that you can read and verify against what you originally meant. These are not ephemeral. They persist. - 04
Claude Code reads them. Every session.
The intent you expressed in step one is in the room every time Claude Code opens a file. Not inferred. Not reconstructed from memory. Present.
What Ada produces
CLAUDE.md The project's semantic document. Claude Code reads this automatically at the start of every session. It contains what the project is, what it is not, and what constraints apply.
agents/ Specialized agent definitions scoped to bounded contexts within your project. Each agent knows its domain.
hooks/ Pre-tool guards that run before Claude Code takes actions. They enforce the constraints from your original intent at the boundary where decisions are made.
These are plain files. You can read them, edit them, version them. Ada produces them. Claude Code uses them.
Try it