Agent Startups: Build Skills, Not Harnesses
Source: https://x.com (tweet) Author: unknown Date: 2025-2026
Summary
Strategic claim: most agent startups are better off building skills rather than harnesses. Big labs will own the harness layer; models are fungible but harnesses are not. The harness layer becomes the new big-lab battleground.
Key Claims
- Big labs will own the harness layer — it becomes a first-party feature
- Models are fungible (can be swapped), harnesses are not (deeply integrated)
- Agent startups should build skills (domain capabilities), not harnesses (orchestration)
- The harness layer = “new big labs battleground” — labs will compete on harness quality
- Skills can survive model replacement; harnesses compete directly with labs
Connection to Other Sources
Connects to Swyx’s Agent Labs Thesis and Anthropic’s effective harnesses. Contradicts or nuances the “full-stack agent company wins” thesis — if you build the harness, you’re on the labs’ turf.
Concepts
- Coding Agents — strategic architecture for agent companies