Dear subscribers,

Today, I want to share a new episode with Kun Chen.

Kun is an ex-L8 principal engineer at Meta who now ships up to 40 PRs a day while rarely reviewing code. In our episode, he walked through the free tools he built to make that possible: Lavish for visual planning in HTML, Treehouse for parallel agents, and No Mistakes for catching AI’s errors before they make it to production.

Watch now on YouTube, Apple, and Spotify.

Kun and I talked about:

I’m proud to partner with Linear

As teams ship faster with AI agents, the bottleneck shifts from building to deciding what to build.

Linear’s new Agent understands your full workspace context: roadmap, issues, customer requests, and code. Ask it to surface patterns across feedback, scope out a spec, or catch you up on team progress.

Try Linear for Free

Top 10 takeaways I learned from this episode

With agentic engineering, you need to spend most of your time in planning and validation. You also need to build a system to manage multiple agents.

  1. Code faster with a team of AI agents. Kun sees himself as the manager of an always-on engineering team. His job is to create plans, validate work, and improve the overall system. Specifically, he runs the same three phases each time:

    1. Plan: Building clear plans is now the most important phase.

    2. Implement: Coding is now handled almost entirely by agents.

    3. Validate: Agents check the work first and only escalate to Kun if needed.

Plan: Spend most of your time here

  1. Plan quality determines how long agents run on their own. A one-line prompt might get the agent to work for a few minutes while a detailed plan can keep it working for hours. To delegate more work to agents, move up these three planning levels:

    1. Prompt: Just explain the next thing to do.

    2. Spec: Get the agent to write a full spec before building.

    3. Goal: Give the agent a clear target to keep trying until it succeeds.

  1. Use Lavish to turn plans into visual HTML artifacts. Lavish is Kun’s free open-source planning tool that lets agents generate a visual HTML plan instead of a wall of text. You can even highlight specific parts to leave feedback for the agent.

Lavish renders the technical plan as an annotated HTML artifact. You can point at any element and leave feedback for the agents.

Implement: Hand the work to the agents

  1. Use Treehouse to run several agents at once. If two agents touch the same code, they might cause merge conflicts. You can use git worktrees to give each agent a copy of the code base, but these worktrees are a hassle to manage. With Kun’s free Treehouse, you can get dropped into an isolated workspace with one command.

  2. Use subagents so the main agent doesn't get overloaded. Kun spins up subagents when a task needs a lot of digging that the main session does not need to remember later. For example, one subagent can inspect part of the codebase while another can test different experiment ideas. The simple rule:

If the work is exploratory, parallel, or likely to fill the context window, delegate it to a subagent and ask it to report back to the main agent with a summary.

Validate: Let agents review the work

  1. Have a different agent review the code than the one who wrote it. Kun no longer reviews all the code that AI generates. Instead he uses No Mistakes, his free validation tool, to run a fresh agent review complete with tests and a risk level. This matters because the agent that wrote the code is biased by its own work.

In Kun’s own testing across 267 agent changes in 15 repos, No Mistakes caught and fixed 68% of mistakes that would have been missed

  1. Only review code the agent marks as high risk. That is the key idea behind Kun’s statement: “I don’t review code anymore.” Kun is not skipping review entirely but instead delegating the first pass to the AI. He only takes a closer look if No Mistakes flags a concern.

How to become great at agentic engineering

  1. Turn anything manual into an agentic workflow. Kun handed off the boring parts of engineering to agents: Testing, docs, naming branches, opening PRs, and checking evidence. Just tell the agent what you used to do by hand and it can work out the rest.

  2. Run /insights in Claude Code to audit your own workflow. Not many people know about this one. The /insights command reviews your past Claude Code sessions and suggest skills, memory changes, and workflow improvements.

  3. Build tools to solve your own problems. To get good at agentic engineering, build a lot of small projects, even if you end up throwing those away, because each one teaches you something. When the same problem keeps coming up and nothing fixes it, build your own tool, like he did with Lavish, Treehouse, and No Mistakes.

Watch the episode and subscribe if you enjoyed the interview.