rauchg: every successful agent needs brain, hands and files — decouple them in the cloud
cramforce · x · 2026-09-24
Vercel CEO Guillermo Rauch argues every successful agent (Muse, Instinct, OpenClaw, Claude Code) has three key components:
- 🧠 Brain: model + harness (logic)
- 👐 Hands: tools, computer, browser
- 🗃️ Files: memories, skills, repos
The easy route is one stateful machine (a Mac Mini running claude or fx all day with caffeine, with storage and CLIs installed). But to run agents cost-efficiently in the cloud you must split them:
- Run the harness on Fluid compute, with durable event logs via Workflow for reliability across restarts, rollouts and crashes
- Hands can be a dedicated browser fleet (Browserbase/Kernel), a Sandbox computer, or lightweight tools like just-bash
- Files: what was missing was decoupled storage — he teases a new capability (thread truncated) for scenarios like memory consolidation
More from coding & agent
- TRACES: A New Benchmark That Grades AI Problem-Solving Process, Not Just Correct Answers — dr_cintas · 2026-09-24
- Anthropic weighs killing Claude Code plan mode, repurposing Shift+Tab for effort levels — benstein · 2026-09-24
- Dev reverse-engineers Qwen Image 2.1 PE, ships ComfyUI node that auto-computes dimensions — BleynSpecnaz · 2026-09-24
- Cua Releases Cua-S1-4B, First Multimodal Decision Model RL-Trained on Live Computer-Use Tasks — multimodalart · 2026-09-24
- Applied Compute uses Jev to auto-cluster failure modes in RL training traces — rhythmrg · 2026-09-24
- Four Steps to Stop Opus 5.5 from Wasting Tokens, From a Real User — dr_cintas · 2026-09-24