Harvard Paper: Coding Agents Re-read 94.8% of Tokens, Pushing Energy Use 25x Higher
rohanpaul_ai · x · 2026-11-18
A Harvard paper analyzing 1 billion tokens from 10 weeks of real AI-assisted development (42 sessions, 752 task episodes) finds cached-context reads account for 94.8% of all API-reported tokens — agents spend most compute re-reading history, not doing new work. Every generated token triggers 260 cached-token reads. Extending an analytical estimator to include cached-context processing yields 250 kWh for the workload vs 10 kWh under output-based EcoLogits, a 25-fold gap. Agent energy accounting must move beyond per-request token estimates.
More from coding & agent
- 1,000+ subagents, 100 hours: GPT-6.2 Astra rebuilds all of Singapore at 1:1 scale — gabrielchua · 2026-11-18
- FastMCP 4.2 ships Skills over MCP — drag a skill folder, get an MCP server instantly — HanchungLee · 2026-11-18
- Amazon paper: weak LLM critics can sink search loops, ideal gain only 0.005 AUC — rohanpaul_ai · 2026-11-18
- OpenAI engineers secretly use Astra for Law to write PRDs and design docs — sherwinwu · 2026-11-18
- START-HERE.md: Mapping Your Vault So Agents Find Files Without Burning Tokens — tomcrawshaw01 · 2026-11-18
- AnyMachine turns a single painting into a 160-part walking robot rig built by an agent in Blender — bennash · 2026-11-18