Google Paper: Replace Agent Chat History With Explicit State, Cut Tokens 16x
rohanpaul_ai · x · 2026-08-29
A new Google paper, SKILL.state, argues long-horizon agents need far less conversation history than we give them: swapping the ever-growing transcript for a small explicit execution state cuts tokens and often improves accuracy.
At each step the model receives only skill instructions, a structured state, and the latest observation; reasoning traces are discarded after producing a validated state update, keeping prompt size roughly constant. On a 100-step warehouse task, Gemini-3-Flash with SKILL.state scored 0.94 using 65,408 tokens vs 0.91 and 1,062,387 tokens for a LangGraph-style baseline — a 16.2x difference. Boundary: everything needed later must fit in the state schema.
Related event: Google Paper: Explicit State Cuts Agent Tokens 16x(2 posts)→
More from coding & agent
- PRAXIST coordinates AI agents for complex tasks like rocket landing — iamfakhrealam · 2026-08-29
- fbtee 4.0 released, fully rewritten in Rust with Oxc — cnakazawa · 2026-08-29
- OpenAI cuts off Cursor, Anthropic cuts off Windsurf, revealing vendor lock-in risks — Yuchenj_UW · 2026-08-29
- Together AI: cascading GLM-5.3 Flash to GLM-5.3 cuts cost 57% while boosting DeepSWE to 80.9% — togethercompute · 2026-08-29
- Seeking open-source methods to extract tables from Indian bank PDFs — OmPatel110 · 2026-08-29
- One Prompt Turns Gemini Flash Into an Optimization Machine: 5000x Gains in 20 Minutes — doodlestein · 2026-08-29