Cloudflare's weekend-trained hybrid model cuts agent cost from 40K to 1.8K tokens
TheTuringPost · x · 2026-11-12
At AI Engineer, Cloudflare's Michelle Chen told the story behind the company's AI neolab: in 2023 Cloudflare put GPUs close to users betting models would shrink, but frontier models exploded instead. When a fast decision-making model appeared, network proximity mattered again — but its weights were closed. After a week of debate, the team trained their own hybrid model Clef over a weekend.
The hybrid routes a CI-triage agent's questions:
- Which of 40 tools do we need? Clef Flash (System 1) picks them
- What's the error in the logs? Clairic (System 1) extracts it
- Why did the build fail? Only this goes to the large reasoning model
Result: same diagnosis with 1 LLM call instead of 6, 1,800 tokens instead of 40,000. The author expects many more System 1+System 2 hybrids, and argues causal/neuro-symbolic methods are needed to verify proposed fixes.
More from coding & agent
- Hackathon project lets AI agents spend company money, but only within CFO-approved limits — kleffew94 · 2026-11-12
- In real orgs the model is rarely the problem — the context it gets is, says Salesforce engineer — msrivastav13 · 2026-11-12
- Salesforce Winter '27 adds AiAgentDefinition metadata for versioned agent deployments — msrivastav13 · 2026-11-12
- Solo dev builds zombie horde level with Tripo P2.0, Fable and new Kaigen engine — gdechichi · 2026-11-12
- United Airlines to Ban Personal AI Agents From Booking Flights for Customers — Polymarket · 2026-11-12
- Agentic Search explained: a 4-search loop costs about 5x one search — and queries leak to more companies — adnan_hashmi · 2026-11-12