Skip to content

Coding With AI in the New Age

The autonomy ladder

A free, technical, bite-sized wiki for people who are new to software engineering but want to build real things with AI coding agents — and not get burned doing it.

AI wrote a lot of the words here; the lessons came from real projects. Everything is grounded in shipped work, official documentation and public security research. Where a page states a number that changes often, it says so.

Two ways to read this. With full-text search, dark mode and a per-page table of contents: vibe.bongwe.space. On GitHub, for anyone who lands on the wiki tab: this page. Both are built from the same lessons, published from the same source.

Who this is for

If you are... Start here
Curious, never shipped anything Set up your workbench → Foundations → Capstone 1
A junior developer using Copilot or chat Agentic workflows → Checks that matter → Safety & security
Building an AI product Prompting & context → System design → Evals → Costs
Leading a small team Rules of operation → Review of AI code → Case studies

The journey

Each lesson is 600–1000 words with one exercise, a diagram, the mistakes people actually make, and links to go deeper. Read one, run one command.

# Section What you get
0 Start Here What AI coding actually is, what the tools are, and a workbench that works.
1 Foundations A correct mental model: models, tokens, context, the agent loop, and where agents are weak.
2 Prompting & Context Instructions and context an agent can actually execute — including the rules file loaded every session.
3 Agentic Workflows The daily craft: plan, implement, test, review, and keep diffs small enough to read.
4 Agent Skills Package know-how so any agent gets it right without being re-taught.
5 Git & CI/CD Git, the GitHub CLI, workflows, the checks that matter, hardening, releases and deploys.
6 System Design Design before code: boundaries, diagrams, state machines, ADRs, schemas, APIs, scale.
7 Shipping & Deploy Railway, Neon, edge workers, environments, promotion, rollback and cost control.
8 MCP Connect agents to real tools with the Model Context Protocol — and manage the risks that come with it.
9 Safety & Security Secrets, generated-code risk, prompt injection, supply chain, OWASP catalogues, privacy, accountability.
10 Quality Tests, integration and smoke tests, evals for AI features, definition of done, debugging.
11 Documentation Docs as code, READMEs, per-feature docs, lesson banks, runbooks.
12 Rules Of Operation SOPs, ADRs, AGENTS.md, skills as operating rules, versioning and change tracking.
13 Team & Process Issues, boards, reviewing AI-written code, token economics, ownership.
14 Case Studies Four real projects, generalised, with the mistakes that actually happened.
15 Capstones Three capstones, a 30-day plan, and where to keep learning afterwards.
16 Reference Cheat sheets, templates, the mermaid cookbook and the link index.

The five rules of this wiki

  1. Verify with evidence. A green check is not proof anyone read the change.
  2. Small diffs, small steps. Blast radius scales with diff size.
  3. Test the deployed artefact, not just your laptop.
  4. Rules belong in the cheapest layer that works — a rule, a skill, a hook, or a CI gate.
  5. You own every line you commit, whoever wrote it.

Start here

Extras

Contributing and licence

Issues and pull requests are welcome — see Contributing and the house style in STYLE.md. Text is CC BY 4.0; the scripts and tools are MIT. Nothing here is legal, financial or security advice for your specific system: adapt it, and verify it yourself.