Snowflake's CoCo Cuts AI Agent Costs by 33-45% with Three Harness Techniques
StasBekman · x · 2026-08-26
Snowflake AI Research presents improvements to its AI coding agent CoCo and CoWork, achieving 33-45% lower cost per trial in end-to-end benchmarks with comparable or better reliability.
- Problem: In long data-analysis sessions, tool schemas, skill catalogs, raw tool results, and conversation history can be re-read every turn, causing context bloat that increases cost and hampers focus.
- Solutions:
- On-demand tool and skill loading: Retrieve only what the current step needs, removing 79K tokens from processing.
- Two other techniques further reduce context processing.
The work emphasizes that agent cost is not solely a model problem but also a context management issue.
Related event: Snowflake Cuts AI Agent Costs by 33-45%(2 posts)→
More from coding & agent
- How to hook Claude Code or Codex up to iMessage in 3 steps — jacob_posel · 2026-10-16
- Are the fastest GPU kernels now written by AI? Kernels are hyper-verifiable, argues one dev — fleetwood___ · 2026-10-16
- Anthropic lead engineer's 4-question prompt makes agents more ambitious, now a free skill — daniel_mac8 · 2026-10-16
- ChainForge dev shows 'feature map' for orchestrating subagents on large PRs — IanArawjo · 2026-10-16
- aiDotEngineer benchmarks agent orchestration pipelines, staging humans as Foreman and workers — vikvang1 · 2026-10-16
- Developer mass-produces 200 videos in hours using Claude Code + Hyperframes — mcraddock · 2026-10-16