Execution Modes

Aigon uses four modes across two axes: number of agents and level of autonomy. All four modes apply to features. Research supports Drive and Fleet only. Feedback does not use execution modes.


                    One Agent          Multi-Agent
                 ┌──────────────┬──────────────────┐
  Hands-on       │    Drive     │      Fleet       │
                 ├──────────────┼──────────────────┤
  Hands-off      │  Autopilot   │      Swarm       │
                 └──────────────┴──────────────────┘

Drive

One agent, hands-on. You guide the implementation step by step.

Best for: most features, learning the workflow, tight control
Workspace: branch or worktree
Slash command: /aigon:feature-start 07
CLI: aigon feature-start 07

Fleet

Multiple agents, hands-on competition. Agents work in isolated worktrees on the same spec. You evaluate and pick the winner.

Best for: important features where you want the best approach
Workspace: one worktree per agent
Slash command: /afs 07 cc gg cx
CLI: aigon feature-start 07 cc gg cx

Autopilot

One agent, hands-off. The AutoConductor monitors the feature and triggers each stage automatically. The agent implements; optionally a second agent reviews; then the feature closes — no intervention needed.

Best for: well-scoped features with clear acceptance criteria
Workspace: worktree (implementing agent) + optional review session (reviewer agent)
CLI: aigon feature-autonomous-start 07 cc
CLI (with review): aigon feature-autonomous-start 07 cc --review-agent=gg


implement → [review] → close

Swarm

Multiple agents, hands-off. Fleet + AutoConductor — multiple agents compete in parallel worktrees. A designated evaluator compares implementations, writes the winner, and the feature closes automatically.

Best for: maximum coverage with minimal supervision
Workspace: one worktree per agent
CLI: aigon feature-autonomous-start 07 cc gg
CLI (explicit evaluator): aigon feature-autonomous-start 07 cc gg --eval-agent=cc


implement (parallel) → evaluate → close

Mode selection guide

Situation	Mode	Why
Routine feature, clear spec	Drive	Fast, simple
Critical feature, need best approach	Fleet	Compare implementations
Well-defined, want hands-off	Autopilot (solo)	Agent runs to close
Well-defined, want hands-off + second opinion	Autopilot (solo + reviewer)	Implement then review before close
Important + well-defined, want best	Swarm	Multiple agents, auto-eval and close
Exploring a technical question	Research (Drive or Fleet)	Investigation before building

Command surfaces

Each mode works with both CLI and slash commands:

Mode	CLI Context	In-Agent Context
Drive	Fully supported	Fully supported
Fleet	Strong for orchestration	Strong for per-agent execution
Autopilot	Primary entry point	Limited (runs via CLI only)
Swarm	Primary entry point	Limited (runs via CLI only)

Agent status tracking

As agents work, they signal status transitions via aigon agent-status:

Status	Meaning
`implementing`	Agent is writing code
`waiting`	Agent finished, awaiting review
`submitted`	Agent committed code and log

Check status:


aigon status              # All in-progress features
aigon status 31           # One feature, per-agent table