Pay-for-results AI pricing fails Dario's own best example: results take years to audit
krishnan · x · 2026-10-28
The author argues pay-for-results AI pricing only works when outcomes are cheap to audit — which rules out Dario Amodei's flagship example. On Dwarkesh Patel's podcast, Amodei said Claude has suggested restarting a Mac 10 million times (worth maybe a dollar), while advising a pharma team to move an aromatic ring could be worth tens of millions. But that molecule needs years of trials before anyone knows the tip's value — who holds the invoice until Phase 3? Outcome contracts will land first where the baseline sits in a system of record: dispatch agents paid per correctly booked load, maintenance agents paid per unplanned stop avoided. Seats reward usage; outcomes reward results you can count.
More from Venture
- Databricks at $7B ARR growing 80%: Ali Ghodsi's CEO playbook revisited — thedealdirector · 2026-10-28
- OpenRouter team launches o11: one API key for agents to pay for anything online — ycombinator · 2026-10-28
- Anthropic IPO filing: $16.3B H1 revenue, compute yield up to $2.08 per dollar — bigdata · 2026-10-28
- Agent infra as the sure bet: Manus runs 1.5M databases on TiDB, investors say — FinanceYF5 · 2026-10-28
- Creator Launches $25/mo AI Community Where Agents, Not Courses, Set Up Workflows — PrajwalTomar_ · 2026-10-28
- FTX Could Have Repaid Creditors Many Times Over Had It Kept Its Anthropic Stake — pycui64 · 2026-10-28