Skip to content
build-3000

Inspired by the Oxford 3000

The essential concepts for building software with AI coding agents

The Oxford 3000 identified the core vocabulary that carries the English language. build-3000 does the same for software engineering: the smallest, most useful set of concepts that takes you from “I have an idea” to shipping secure, production-ready software - with an AI agent as your implementation partner.

AI writes the code. You stay the director.

Levels
6
Modules live
40
Core terms
200+
Hands-on projects
40

The ladder

Six levels from first idea to production AI applications. All six are live today - the full ladder, foundations to AI-native.

How it works

Reading is not finishing. Every module ends in a built project, verified twice - by you, then by your own AI agent.

  1. 01

    Learn

    Read a focused module: vocabulary, concepts, worked examples and the mistakes everyone makes.

  2. 02

    Build

    Direct your AI agent through the module's real project using the brief and agent context provided.

  3. 03

    Self-verify

    Check your work against the Definition of Done - item by item, with evidence, in an interactive checklist.

  4. 04

    Agent-verify

    Paste the generated verification prompt into your AI agent. It audits the same list independently.

  5. 05

    Compare

    Where your self-check and the agent's audit disagree is exactly where your understanding grows. Then move on.

“AI generated it” is never done

Every module ships with a Definition of Done: functional, architectural, security, testing and production checks that grow stricter as you climb. You verify by hand first. Then the twist -

build-3000 generates a verification prompt for your own AI agent. It audits your project independently, against the same list. Two perspectives, one checklist - and the gaps between them are where real understanding forms.

See a live module →

The dual-check, in practice

Your pass

Restart test performed twice. Data survived both times. Edit never duplicated a row.

Your agent’s audit

FAIL: totals computed from UI state, not stored data. A restart recalculating from storage would expose drift...

The disagreement is the lesson.

Open source, MIT licensed, built in the open

The curriculum lives in JSON files validated by schemas. Found a mistake? Know a better example? Add a module, fix a definition - every contribution goes through the same validation gate the site itself uses.