So while all harnesses make slightly different decisions while performing this "plan, act, verify" pattern, this loop is…
SemiAnalysis Twitter · SemiAnalysis (@SemiAnalysis_) · 2026-07-03
SemiAnalysis concludes their harness explainer by arguing that the 'plan, act, verify' loop is the universal skeleton of all agentic tasks, and that model quality—not harness engineering—is the primary driver of agentic performance.
Appears in
Extraction
Topics: agentic-codingllm-harnessesmodel-capability
Claims
- All agentic coding harnesses share a common 'plan, act, verify' loop as their fundamental operating skeleton.
- Harness engineering provides incremental improvements but is not the primary source of agentic capability.
- The real power in agentic systems comes from the underlying models, not the orchestration layer.
Key quotes
let's stop talking about harnesses like they're magic! While good harness engineering helps, the real power is still in the models themselves. Everything else is just REST all the way down.