A practical guide / AI-assisted development

Build with AI. Keep the work reviewable.

Start with a useful constraint, give the agent the right context, and build a workflow your team can inspect and improve.

Follow the reading path ↓

Start here

Three useful places to begin.

Read in order, or choose the problem closest to your own.

  1. 01

    Start with a reviewable change

    Use a small pull request as the unit of progress. Look at what the agent changed before asking it to take on more.

    The 200-Line Rule That Changed How I Ship Code with AI ↗
  2. 02

    Give the agent a map

    Keep repository guidance discoverable. Separate the rules needed for every task from details the agent should look up.

    Optimizing CLAUDE.md: How to Scale Your Codebase Documentation for AI Assistants ↗
  3. 03

    Add coordination deliberately

    Explore the tradeoffs of multiple agents: task boundaries, shared context, and the review work that remains yours.

    I Built an Autonomous Coding Swarm. Here's What I Learned. ↗

The working principles

Give the agent a smaller, clearer job

A useful starting point is a change you can describe and review. Write down what should happen, where it belongs, and how you will know it works. Ask the agent to inspect the existing code before proposing an implementation.

The 200-Line Rule describes one constraint I use to keep changes manageable. The particular number is less important than the discipline: keep each change small enough that a person can understand its consequences.

Make context discoverable

An agent does not need every detail of the repository on every turn. It needs a reliable place to find the rules, the relevant examples, and the commands that check the result. Keep shared instructions concise and link to deeper material when the task requires it.

Optimizing CLAUDE.md explores that boundary. For a concrete workflow, From AI Assistant to 10x Engineer follows the work from a task to a production-oriented implementation.

Treat verification as part of the work

A convincing explanation is not evidence that a change works. Read the diff, run the checks that exercise the behavior, and try the changed flow. For interfaces, that includes the browser. For business records, check the resulting data and retry behavior using an isolated environment.

Only add more autonomy when you can still inspect what happened. The autonomous coding swarm account covers coordination problems and the tradeoffs of dividing work between agents.

From the idea to the work

Make the workflow useful in a real product.

Bring a product idea, an internal tool, or an AI-assisted workflow that needs a working implementation.

See the Lively Icons case study →

Go further

More notes on ai coding.

Further experiments, implementation details, and lessons from the work.

How I Run 3 Engineering Teams with 1 AI Brain ↗

I manage three engineering teams across two products with an AI system that runs daily standups, triages PRs, monitors production, and manages my inbox. Here is the actual architecture and what it gets wrong.

Claude Code's Source Leaked. Here Are 7 Things I'd Change. ↗

The Claude Code source code became publicly accessible. I spent a day reading it. The engineering is impressive, but there's an architectural blind spot: the system optimizes for content delivery but not attention management. Here are 7 concrete changes I'd make.

Plugins Are Wasting Your Tokens, Thanks Vercel ↗

I traced where my context window was going during a Claude Code session. A single plugin injected 48,000 tokens of irrelevant documentation across 6 turns. Here's what I found and why plugin architecture for AI coding tools needs a rethink.

Leveraging AI for Problem-Solving ↗

AI isn't just for automation. It's a powerful thinking partner for solving complex problems. Here's how to integrate it into your decision-making process.

Browse all writing →