A software artifact is a concrete, structured output that another agent, process, or reviewer can consume. In orchestrated AI workflows, artifacts replace sprawling conversation histories with versioned, traceable, auditable handoffs. section 5.6.5
Useful artifacts create explicit control points across an agent pipeline:
- Design documents carry planning decisions into implementation. section 5.6.5
- Code and decision summaries make implementation outputs inspectable. section 5.6.5
- Test results and coverage analysis support independent verification before documentation begins. section 5.6.5
- Documentation diffs and changelog entries preserve downstream changes. section 5.6.5
Artifact flow in a payment-retry pipeline
An architect emits design.md, implementers produce resilience.py and gateway changes, and testers remain queued until implementation completes. The pipeline view exposes status, timing, token use, and dependencies instead of leaving failures hidden in parallel conversations. section 5.6.2
Pipeline: payment-retry-feature Status: Running (7m 23s) [✓] Architect (2m 14s, 12k tokens) → design.md [✓] Implementer-1 (3m 02s, 28k tokens) → resilience.py [◐] Implementer-2 (4m 11s, running) → gateway changes [ ] Tester-1 (waiting) [ ] Tester-2 (waiting) [ ] Documenter (waiting) Total tokens: 47k ($0.42) Estimated remaining: 35k tokens, 5 minutes
Inspect the pipeline listing to see artifacts, agent status, queue dependencies, duration, and token consumption in one operational view. section 5.6.2




