Try Intent-System→
Product

Humans decide. Agents implement. You review against intent.

Per-slice assignment, not per-project. The intent tree supplies the context regardless of who implements. Humans don't disappear from the loop — they move to a different layer.

Intent-System · Step 5 of 7↑ Overview
  1. 01Current Flow
  2. 023 Threads × 6 Steps
  3. 03Intent Tree
  4. 04Unified Intents
  5. 05Human + AI Assignment
  6. 06Review & Closeout Loop
  7. 07Roadmap

Roles in an Intent-System loop

Intent owner (human)
Curates the intent tree. Makes promotion decisions (inferred → clarified → canonical).
Clarifier (human)
Reviews the batches of "what about this?" questions AI surfaces against the intent tree, with a recommended answer beside each. Accepts the recommendation where they have no strong opinion; writes their own where they do. Each accepted answer is promoted into the tree, so every later issue inherits the new context.
Issue compiler (system or human)
Cuts an issue-ready slice from the intent tree. The resulting issue inherits the relevant architecture, contracts, and UI patterns automatically — which is why writing a great issue, the highest-leverage skill in modern AI-assisted coding, becomes feasible at all.
Implementer (human / agent / AI)
Implements the slice and opens a PR against ai-develop. Per-slice assignment — choose what fits.
Reviewer (AI primary, human final approval)
The reviewer is fundamentally an AI agent that has read the entire intent tree. The implementer works from what is embedded in the issue; the reviewer reads across the implementation, the issue, and the intent tree at the same time and judges whether the code and the tests honor the intent. A failed review is routed back as a clarification, not as a taste comment. A human gives the final approval before promotion.

Three assignment patterns

Fully autonomous agent loop

Agent picks issues, implements, opens PRs continuously. Humans intervene only on review / promotion.

Agent + human review (default)

Agent implements; human reviews every PR before promotion. Most common configuration today.

Human implementer with intent context

A human writes the code; the intent tree is the briefing material. Useful for new engineers or sensitive domains.

What humans still own (and why)

  • Mission / Vision / Value. These define what the system is for; agents can suggest, but humans decide.
  • Clarifications. Ambiguity in the intent is a human-only resolution.
  • Final accept on PRs. Reaching main is a human decision, even when agents reviewed first.
Try Intent-System→