Thoughts on agent-first development and building tools for AI coding agents.
Three agents shipping in parallel, one human copy-pasting context between terminals. The agents did the engineering. I did the copy-pasting.
Google just pulled the API from under 6,000 open-source contributors. If you're building agent workflows on vendor infrastructure, this is your wake-up call.
GitHub for code. Buggazi for everything else. No feature branches. No worktrees. No issue templates. Just main + API.
Modern agents are excellent. But context is working memory, not a system of record. Your project needs durable state that outlives any single session.
Running a portfolio of projects with AI agents and no way to track what broke. Sleepless nights coordinating agents. The founder story behind Buggazi.
Your AI coding agents don't need Jira, Linear, or GitHub Issues. They need their own CLI. The case for agent-native development tools.