The interesting Pi workflows are not just “use an agent.” They are transitions: from TUI to packages, from packages to scripts, from scripts to worktrees or queues, and from raw power to explicit controls.
Most users should start in the TUI and only move into JSON mode, RPC, or SDK embedding once they understand how Pi sessions, context, and extensions actually behave.
Artifacts: pi (pi-mono), pi.dev
Once something is useful twice, the Pi ecosystem tends to turn it into a skill, extension, prompt template, theme, or npm/git package rather than keeping it as a private prompt trick.
Artifacts: pi-skills, agent-stuff (mitsupi), pi-share-hf
A big part of Pi's distinctiveness is that it can stop being a chat app and become infrastructure for bots, scripts, or embedded tooling.
Artifacts: pi (pi-mono), pi-acp, task-factory
When a single session stops scaling, Pi users either wrap it in queue-first review lanes or in shell-native worktree swarms; both paths show up clearly in the ecosystem.
Artifacts: task-factory, PiSwarm
The ecosystem often starts from speed and composability, then layers in notification, audit, sandboxing, policy, or redaction once the workflow becomes important enough to trust.
Artifacts: pi-notify-pp, nono, toolwatch