Aeion Forge for Hybrid Human + AI Teams
Modern engineering teams aren't human-only or AI-only — they're hybrid. Senior engineers writing critical code, AI agents handling refactors + tests + docs, junior engineers learning from AI suggestions. The coordination model needs hand-off boundaries (who does what), trust scopes (what AI is allowed to merge unsupervised), audit trails (who decided what), and pattern propagation (when an AI learns something, share it). Aeion Forge provides the workspace where human + AI teams collaborate with clear boundaries.
What Hybrid Coordination Demands
`
Walkthrough — Senior + AI Pair Programming
`
Walkthrough — AI-Driven CI Bot Team
`
Walkthrough — Junior Developer + AI Mentor
`
FAQ
Forge doesn't ship a permission engine that blocks agents from specific tools or paths — trust scope is a team convention you write into the agent's instructions. What Forge enforces mechanically: file locks (claim_file_lock) to make in-progress work visible, a status workflow (submit_for_review → approve_issue/reject_issue) any human can gate on, and a Definition-of-Done check on complete_issue requiring a linked commit or PR.
Per-action attribution (agentId/sessionId, timestamp). Session timeline + issue history for reconstructing what happened. Reverting bad code is a normal git revert — Forge tracks the issue/decision side, not source control.
Any decision can be superseded (supersede_decision) or updated. Issues route through submit_for_review before approve_issue/reject_issue, and an agent can explicitly pause with request_human_review.
Per-agent ID tagged on comments, decisions, locks, and commits. No built-in reputation score — that judgment stays with your team.
No built-in per-agent PR-acceptance or false-positive rate. You can reconstruct an agent's activity from its attributed comments/decisions/progress logs via get_session_timeline / get_issue_history and compute your own.
Forge has a template system (issue / spec / prompt / workflow types) you can define once and apply repeatedly — handoff steps and trust-scope conventions live in the template content, not as a separate enforced policy object.
attach_commit and comments carry an optional agentId, so you can trace which agent touched what — treat it as an attribution trail, not a formal legal-provenance system.
No built-in PR-merge-rate or bug-introduction-rate metric. Track quality through your normal review/CI signals; Forge's job is keeping the issue/decision/knowledge trail those signals point back to.
Per-team convention, written into agent prompts and enforced through your review process — not a Forge-side configurable policy object.
No automatic trust-score or graduation — the senior decides when to loosen the review gate, using Forge's issue/PR history as the evidence. Forge tracks the trail; the senior makes the call.
Per-incident attribution. Per-incident root-cause analysis. Per-incident process-improvement.
Per-new-hire AI mentor. Per-new-hire knowledge-base curated. Per-new-hire progress tracking.