Appearance
Level 4: Team Development with Coding Agents
Last updated: 2026-07-27
A 10x individual contributor with AI is impressive. A team of them, without shared practices, is chaos.
The previous chapters focused on individual mastery — how you prompt, understand, and orchestrate coding agents. But software is a team sport. When every developer on a team is generating code at 5-10x the previous rate, the bottleneck shifts from writing code to coordinating, reviewing, and maintaining it. This chapter addresses that shift head-on.
This chapter is still a draft. Its structure follows the lifecycle of team-scale AI development:
The foundation: context — Context Management Principles explains why stuffing everything into a large context window actually hurts performance. This chapter scales that individual discipline into shared team practice.
What to build — Specs as Source of Truth argues that when AI can generate code faster than humans can review it, the specification becomes the real artifact. Specs are no longer documentation — they're the source code of intent. This section covers how to write specs that agents can consume and humans can verify.
How to share context — Team-Level Context Engineering scales the context management principles from Chapter 2 to an entire team. It covers shared AGENTS.md files, layered context hierarchies, and the four optimization targets (correctness, completeness, low noise, trajectory) that determine whether your team's agents produce coherent output or contradictory spaghetti.
How to stay lean — Intentional Compaction introduces a discipline most teams skip: continuously summarizing and distilling work state into structured artifacts. Without this, context windows fill with stale information and agent performance degrades over time.
How to build workflows — Design Agentic Workflows in Daily Work moves from one-off prompting to repeatable team workflows, including agent ensembles, verification loops, and lifecycle integration.
How to review — Code Review at AI Scale reimagines code review for a world where PRs arrive faster than humans can read them. The new goal isn't line-by-line inspection — it's keeping the team aligned on what's changing and why, with AI handling the mechanical checks.
How to stay safe — AI Testing, Security & Failure Modes addresses the dual transformation: AI introduces new attack surfaces while also offering new defensive capabilities.
A chapter on security and permission boundaries is planned but not yet written. See ROADMAP.md for what's outstanding in this level.
Where you've been: Chapter 2 made you dangerous as an individual — you know the machine and its full toolkit. This chapter ensures your team doesn't implode under the weight of AI-accelerated output.
What's next: Chapter 5 zooms out from daily practice to the bigger picture. What does "Software 3.0" actually mean? What mindset do architects need in a world where code is cheap and judgment is scarce? And where is this all heading?