Product Manager AI Workflow

Built for controlled AI handoffs.

Most AI product work breaks down when a single chat has to carry every source document, every draft, every correction, and every execution decision.

Keep the main thread light

Codex holds product intent and decisions while worker lanes handle heavier context.

Make worker scope explicit

Kiro and DeepSeek receive bounded tasks with clear files, goals, and permissions.

Review before continuing

The workflow is intentionally not an endless loop. One useful step, then review.

Try it in 3 minutes.

Generate a worker prompt first. It does not call Kiro or DeepSeek and does not spend credits.

git clone https://github.com/qianchongyang/kiro-ai-workflow.git
cd kiro-ai-workflow
python3 ai-workflow-cli/cli.py --help

python3 ai-workflow-cli/cli.py prompt \
  --type T4 \
  --goal "Extract user roles, workflows, risks, and PRD open questions" \
  --files "Customer notes: users need batch messages, approval, retries, and statistics."

What comes next.

The first release is usable. The next work is about packaging, repeatable sessions, and better reporting.

Session workflow

Commands such as session start, session next, and session review for continuous projects.

Ledger export

Markdown and CSV exports for cost reporting, handoff records, and team review.

More examples

Packaged examples for PRDs, HTML prototypes, development handoff, and review tasks.