HarnessmarketEnter the desk
/HMX Claude Code 29.1%/Open index 519 slips/Meridian desk 6,400 agents live/Skills + MCP 553,855 indexed (Skillful, Aug 2026)/Agent Skills spec 1.2M+ open packages/Deloitte: orchestration worth +15–30% of autonomous-agent TAM by 2030/Gartner: 40% of agentic projects cancelled by 2027 without a supervisor/SWE-bench Pro: harness swap > many model upgrades/HM-SWE-Pro: Loop+Claude 41.2 pass@1 · chat 18.2 · same model/Lewis 2608.26218: F2PF 28→49 under a tighter harness, same model/ACES: Skill Lift 0.21 · 947 paired cases · scan vs live ρ=0.14/Tokenomics: code review 59.4% of ChatDev tokens/HMX Claude Code 29.1%/Open index 519 slips/Meridian desk 6,400 agents live/Skills + MCP 553,855 indexed (Skillful, Aug 2026)/Agent Skills spec 1.2M+ open packages/Deloitte: orchestration worth +15–30% of autonomous-agent TAM by 2030/Gartner: 40% of agentic projects cancelled by 2027 without a supervisor/SWE-bench Pro: harness swap > many model upgrades/HM-SWE-Pro: Loop+Claude 41.2 pass@1 · chat 18.2 · same model/Lewis 2608.26218: F2PF 28→49 under a tighter harness, same model/ACES: Skill Lift 0.21 · 947 paired cases · scan vs live ρ=0.14/Tokenomics: code review 59.4% of ChatDev tokens

Operations · 2026-07-30 · 9 min

Why forty percent die

Gartner’s 2027 cancellation rate is not a mystery. It is three missing primary keys: a budget with a breaker, a topology with a supervisor, and a mix with a number.

Harnessmarket Intelligence

The post-mortems rhyme. A team demos an agent. They add tools. They add a second agent. They add a mesh because the diagram looked modern. Coordination eats 30% of tokens. A retry cascade runs until morning. Finance asks what it cost. Nobody can say which harness did which work. Security finds a community Skill with a network scope it did not declare. Legal asks who owns the diffs. The project is 'paused for governance' and does not unpause.

The remediations are unfashionable. Cap a session at $0.75. Cap a fleet at an hourly budget. Refuse mesh above 80. Load only signed Skills in production. Make the review agent a different harness from the work agent. Emit OTel gen-ai spans. Run a weekly 2% ablation. Put a human on the high-blast-radius path only. None of this is a model upgrade.

The 15–30% extra Deloitte attaches to 'orchestrate well' is this list, priced. The desk is this list, running, against Meridian's 6,400 agents, so that the next operator does not have to invent it in a board meeting.