Product Direction

Operator correction

The platform should not become a full IDE.

It should be a connector/control plane that sits next to existing coding-agent products:

• Claude Code
• Codex
• Hermes
• OpenCode / Copilot later

Best terminology

Operator wording	Calibrated term
“Claude Code connector or Codex connector”	coding-agent runtime adapter
“visual on the left-hand side”	workflow telemetry panel / run visualizer
“n-flow of some sort”	node-based workflow visualization, similar to n8n/River/Temporal-style execution graph
“how the harness is working”	artifact-gated workflow observability
“not an IDE”	control plane, not editing surface

Product category

• Standard category: Internal Developer Platform / developer portal / control plane.
• Emerging category: Agentic SDLC platform / coding-agent orchestration layer.
• Internal term: Universal Agentic SDLC Harness.

User experience

User opens Codex or Claude Code
→ connector injects/generated project harness context
→ user asks agent to fix/build/research
→ harness creates task + run packet
→ Vercel UI shows the visual flow
→ each stage writes artifacts
→ required gates block completion if missing
→ final answer includes proof paths and human decisions

What we are building first

A Vercel-hosted run visualizer and commissioning surface, backed by local/CLI runner adapters.

The first release should prove:

1. The platform can commission a repo.
2. It can generate a project-specific harness pack.
3. It can launch or wrap an external agent runtime.
4. It can stream run/stage/gate status.
5. It can block “done” when required artifacts are missing.

What we are not building first

• Browser-based code editor.
• Full VS Code replacement.
• Private HumanLayer app clone.
• Hosted remote runner before local runner works.
• Enterprise SSO/audit before task/artifact/gate model works.