Workflow uses Jev routing to cut Opus 5.5 call costs
A workflow puts Jev before expensive Opus calls to decide which tasks can reuse a verified result, which need simple routing, and which need Opus to investigate, with spending caps and retry limits.
an Anthropic engineer just leaked a confidential internal guide to cutting Opus 5.5 costs with Jev 9-minute read, a complete workflow you can adapt to your own system give it your recurring tasks, saved results, and quality checks, then put Jev before the expensive Opus calls → what can reuse a verified result → what needs a simple routing decision → what actually needs Opus to investigate the workflow checks saved results and input versions before another paid call, so an unchanged task doesn’t trigger the same expensive work again when Opus is needed, it gets a specific task, the relevant evidence, and a clear check for what counts as finished if that check fails, the next attempt gets the failure and its evidence, fixing one problem instead of restarting the whole job your priorities, spending cap, and retry limit are part of the setup, with code enforcing the boundaries before more work runs start with one task you repeat every week, track the calls and cost, then expand the workflow once the results hold up save this, then read the full Opus 5.5 + Jev guide below for the routing, saved results, and targeted retries ↓
This post quotes another post: x.com/0xnineinch/status/2108604801411879139