HAR is an open source harness for coordinating multi-agent coding workflows, posted on Product Hunt with a one-line description and no public docs yet.
The pitch targets a specific gap in current AI coding tooling:
- Multi-agent orchestration — routing tasks across specialized coding agents instead of one model doing everything
- Open source — no vendor lock-in, forkable for custom workflows
- Workflow harness — framework-level coordination, not a chat interface
Most AI coding tools in 2025 still operate as single-agent systems. Cursor, Claude Code, GitHub Copilot — each runs one assistant per session. Multi-agent frameworks exist (CrewAI, AutoGen, LangGraph) but they’re general-purpose orchestration layers, not built specifically for code generation, review, and test-writing loops.
NOTE
HAR’s Product Hunt listing doesn’t yet include a GitHub link, installation instructions, or architecture docs. This is an early-stage project to track, not something to deploy in production today.
The concept lands at a moment when single-agent coding is hitting real limits. Andrej Karpathy’s “vibe coding” framing assumed one human directing one model. Multi-agent coding pushes further — a team of agents, each with a role, coordinated by a harness layer that manages context, handoffs, and conflict resolution between them.
Worth bookmarking if you’ve outgrown single-agent coding workflows — but verify repo maturity and actual feature surface before building anything on it.