🎖️ AI Readiness Index

Domain Driven Design is the best thing you can do when building codebases with AI.

PRs get simpler, surface area touched per prompt gets smaller, and multiple agents can work without really introducing conflicts.

I came up with a metric — the AI Readiness Index ^{(patent pending)} — a simple way to gauge whether an organization is actually AI-ready. Ask four questions:

1. Is your data model domain-isolated and well bounded?
2. Is the system documented in a machine-navigable way?
3. Is your data queryable, discoverable, and semantically described?
4. Do you have a proper interface/tooling layer (MCP or equivalent)?

Naming matters more than people think

Even simple things — naming — matter more than people think.

uid tells neither humans nor models anything. platform_user_id immediately signals ownership and scope.

Same with:

1. ext_id vs external_platform_id
2. cfgvs billing_configuration
3. ts vs created_timestamp

One table, three domains

The problem goes deeper. A single users table with billing, auth, and platform context is three domains in one table. A billing user might not even mean the same thing as the auth user.

Now, to build an agent on top of it, you end up inventing RLS po and all sorts of accidental complexity — when it could havesimply been avoided by splitting it into three schemas, each with its own actions, verbs, domains, and simple policies.

Go back to basics

Go back to basics. You can be way faster and much more predictable.

Honorable mentions: the Ash framework in Elixir