Jev in Practice: Browser Use and Agent Workflows

The DepthFirst team applies Jev to binary vulnerability analysis and discusses pairing GPT-6 Sol for threat modeling with Jev for fast scanning.

Running the same target twice with only the Jev decision layer toggled, the agent reported 18 findings versus 16, recalibrated 9 conclusions, finished faster, and kept decision-layer cost under $5.

The author built a meme trend monitoring agent with CREAO and found most of its tokens went into deciding what to do next, which frames Jev's approach of handling frequent small decisions separately.

A Taiwan finance podcast touches on Jev during a semiconductor discussion, saying it suits clearly bounded routing and gating but not financial report analysis or complex workflows.

Diogo Amogo proposes moving context decisions out of the expensive LLM loop to make coding agents faster and cheaper, with 10 key ideas listed in the post.

The author estimates Jev's Doom demo makes about 10 decisions per second at roughly $7 per hour, which would cost $168,000 a day for 1,000 players online, so he has scripts handle routine behavior and lets Jev decide only at key moments.

Opus 5.5 reads the dispute and Jev checks the reasoning against tracking records in 40ms to decide whether PayPal releases funds, blocking 88 refunds a day that should not have gone out.

Industry practitioners say Jev is essentially a fixed-choice classifier, scoring 83.2% on banking77 versus 93.3% for bge-small plus logistic regression, and that its "can't hallucinate" claim only guarantees valid format.

A no-screenshot loop and single-round-trip design cut protocol calls from 1,092 to 101, completing a Google Flights search in 7.1 seconds.

AY Automate retested with 791 labeled decisions and found Jev's median latency at 0.33s versus Terra's 1.17s, only 3.6x faster; routing the confident majority to Jev and the uncertain 19-23% to Terra matched accuracy at roughly 26-28% of the cost.

Presented at Agents That Act: a TrueFoundry agent harness runs reproductions while Jev checks results against the paper's claims, reducing reviewer workload.

Across 44 benchmarks and 7,193 responses, Jev's single-question judgments reached a median AUROC of 0.886, with evaluation costing $0.30 versus $18.96 for LLM judges.