The Compute Economy of AI Agents
summary
Agents delegate work to specialist services, creating a real marketplace where routing choices determine costs and winners. Concrete numbers: 28 million active agents today, projected billions by 2030; agent pipelines cost ~ $0.05–$0.51 per execution; GPT-3.5 Nano at $0.05 per million input tokens versus Cloud Opus $5 input / $25 output per million. Examples: Klarna cuts contact volume and saves $40M; IBM reduces false alerts 40% and speeds incidents 30%. Protocols MCP and A2A plus on‑chain settlement (USDC, X402) enable machine payments.
AI-written summary · about
chapters (11)
- 0:00The delegation economy
- 2:01From monolith to modular
- 5:43Real world agent deployment
- 8:36The compute marketplace
- 10:09Unit economics of agents
- 14:23Speed vs. accuracy trade-offs
- 16:01Agentic payment systems
- 18:10Standardizing protocols
- 22:03Challenges to production
- 24:50Who wins in the agent era
- 28:46Building for non-human users
related talks
45mFrom Tools to Partners: The Rise of Agentic AI
57mWhen AI Agent Teams Work (And When They Just Create More Work)
29mWhat exactly are AI Agents?
27mFrom Theory to Real-World Impact
28mAI and the state of APIs in 2026
15mHow AI is transforming possibilities into reality
27mSoftware engineering with GenAI
14mReliable Agentic AI, the Open Way