💻 Tom Tunguz: The AI Harness Now Matters More Than the Model
Tunguz argues the harness — the scaffolding wrapped around a model — has become the faster-moving variable in AI coding cost and quality.
• OpenAI's GPT-5.5 scored 61.5% functional correctness in its native Codex harness, but 87.2% in Cursor's harness — a 25.7-point swing.
• Anthropic's Opus 4.7 also scored higher in a rival's harness (91.1% in Cursor) than its own (87.2% in Claude Code).
• The harness now "owns the prefix," accounting for 86-98% of the total bill.
💡 Why it matters: If the harness drives more of the performance gap than the underlying model, buyers and builders may need to re-evaluate vendor lock-in and tooling choices as much as model selection — a shift in where competitive advantage in AI coding actually sits.
🔗 Read more
#tomtunguz