Abacus AI's Jev bets on typed decisions: one forward pass to let agents think in parallel

bindureddy · x · 2026-09-17

Bindu Reddy, CEO of Abacus AI, argues the smartest idea this week isn't a bigger model but Jev: a small model that returns a decision instead of text—one forward pass, one probability per option, no tokens to wait for.

Her case: today's agents are painfully serial—pick a tool, wait; pick a branch, wait; retry or abort, wait. With typed decisions, an agent can fan out 50 sub-tasks and score them all at once, running planning, verification, and safety checks in parallel. The LLM stops narrating and starts scheduling.

Her thesis: chain of thought made agents deeper; typed decisions make them wider. Winning agents won't think harder about one step—they'll think about a thousand steps at the same time. The feature is shipping in the Abacus AI Agent.

Related event: TypeSafe exits stealth with decision model Jev and RLCD training method(76 posts)→

Original post →

More from coding & agent

coding & agent channel →