CHAPTER 14 · Autonomous Task Agents: Manus and CodeAct · 6 / 10
Why it matters
Manus shows what the harness ideas look like when pushed to a fully autonomous, general-purpose agent. CodeAct is a genuinely different action model that you may want for agents doing varied, open-ended work (it trades a bigger action space for tighter sandboxing needs). And file-based memory, especially the todo.md pattern, is a cheap, robust technique any agent can adopt to survive long tasks without drowning its context window. Together they round out your toolkit beyond the repo-focused coding agents.