Agent bottlenecks are in the loop, not the model, cutting costs by 30%
Background-Job-862 · reddit · 2026-08-27
Comparing agent runtimes with a fixed Claude Opus model and identical success rates (11/14 tasks), the author found massive performance gaps: the fastest runtime was 2.5x quicker (39 vs 96 mins), used 3x fewer tokens (3.85M vs 13M), and cost 30% less than the slowest.
The differences stemmed from "boring runtime stuff": how much system prompt/tool context is resent per turn, how tool outputs accumulate, and the aggressiveness of loop exploration/retries. This suggests agent benchmarks measure the combination of model + harness + prompting + tool loop, not just the model.
More from coding & agent
- Musk amplifies dev's verdict: Grok Bot easier to use than Codex and Claude — elonmusk · 2026-10-15
- Blob0 launches: a macOS file-manager-style desktop client for S3 and Cloudflare R2 — yihui_indie · 2026-10-15
- The 6-Layer AI Engineer Skill Stack for 2026: Vectors to Multi-Agent Ops — mdancho84 · 2026-10-15
- Building a Mechanical Engineering Agent Workflow in ChatGPT: 30-Node Canvas Shows Promise — burhop · 2026-10-15
- Open-source Harness Engineering course hits 19k stars, shows $9 vs $200 gap between failing and working agent setups — thisguyknowsai · 2026-10-15
- A 6-person team shipped healthcare prior-auth AI agents in 3 months — here's the playbook — alex_verem · 2026-10-15