100B decisions in 5 minutes: a layered agent pipeline where Opus only probes 96 hotspots
Arindam_1729 · x · 2026-10-07
Arindam showcases a layered decision system built with TypeSafe's JEV, Opus 5.5, and worker agents, designed so the frontier model never sees the full stream.
- JEV is the brain: 5,800 nodes running continuously, making 100B decisions in 5 minutes (3ns each)
- Opus 5.5 is the spider: it walks the graph, finds hotspots, and probes only 96 nodes
- Dots are the workers: Echo, Sprout, Pocket and Stella ingest, train, store and rank, yielding 31 deliveries
The full pipeline: 100B decisions → 96 probes → 31 deliveries → 1 screen. The author contrasts this with the usual "input → biggest model → answer": a cheap fast layer finds what matters, the expensive model investigates only hotspots, and agents act. Includes a linked article on building projects with JEV.
More from coding & agent
- Prompt2ELF: an AI model as the front end of a zero-toolchain software pipeline — UncommonSort · 2026-10-07
- Beating 5-hour limits: a multi-model AI coding workflow split by task value — AJChadha · 2026-10-07
- 262k-star galaxy collision simulator built with Opus 5.5, Three.js and Blender — techartist_ · 2026-10-07
- Developer resumes robot lamp project using OpenAI Codex — kamathsblog · 2026-10-07
- Pre-registered study: selection beats extraction in agent memory, 3061x cheaper to write — Rishabh Sharma · 2026-10-07
- Cursor iOS app adds remote control for local coding agents, on by default — tetsuoai · 2026-10-07