The processes, tools and patterns that work well for human software teams translate reasonably well to agent-first teams.
Managing many chat threads across models like Claude and Codex eventually breaks down, and chances are you, the human, are the bottleneck. I noticed my conversations in each thread drifting toward excessive shortcuts, too-quick thinking and occasional confusion as I crossed between them.
The processes I've championed my whole career are still critical in agentic development. Planning and design feed a better implementation plan, which feeds a better testing plan. Design, Implementation and Testing collaborating together still matters, whether the contributors are people or agents. So I built the same boards my human teams worked from, and taught my agents to respect the workflow whether the work looks like a quick bug fix or a heavy feature.
Communication stays transparent: every discussion, design revision, problem and outcome is stored and available to future work. The same processes, tools and philosophies that helped some human teams get better results are helping agent teams reach them too.