There are at least four fundamental engagement patterns when working with AI.
The main differences are:
- context scope: what information is kept and for how long
- surface: tool/paradigm AI and human use to interact and collaborate
| Mode | Surface | Example Tooling | |
|---|---|---|---|
| Lookup | one-off questions: user sends question, agent answers. no context kept between questions. | messenger, no threads | NanoClaw |
| Workshop | ephemeral sessions: user and agent work together toward a goal. context is shared within the session, discarded when it ends. | web UI + coding agent + git repo + git worktrees | Clay sessions/worktrees based on GitHub issues |
| Companion | persistent sessions: ongoing collaboration with context that accumulates across sessions. | web ui + coding agent + git repo | Clay |
| Mission | autonomous task sessions: agent takes a task definition incl. definition of done, works independently, only asks user for approvals. | GitHub issues + PRs + Kanban | ? |
Notes
- I think there is a pattern beyond “Mission” that adds another strategic layer on top, examples are Linear.app and Dan Meisner’s “Mission Control”. I’m not sure yet if that is an extension of “Mission” or something separate.