Research & models Nov 16, 2026
A group of authors reports in their paper's abstract that adding process control outside the model, without changing the underlying model, raised the unattended completion rate on a 120-case benchmark the authors built from their own business workflows from 50% to 100%.

Companies building their own AI agent platforms get one deployer's self-reported data point on whether a stronger model or stronger control outside the model matters more. The authors' code sits under the ZTE AICloud account of ZTE, the Chinese telecom-equipment maker, and the tests cover the authors' own deployed workflows, each longer than 200 steps. The method moves the workflow definition, state and domain knowledge out of the model into versioned files, and at run time a state machine and machine-checkable acceptance criteria guard each step. On the same benchmark, end-to-end latency fell from 440 minutes to 159. The abstract doesn't say what business the workflows serve, what baseline setup the 50% figure came from, or whether the 120 cases overlap with the cases used to tune the workflow (arXiv 2611.17696, 2026-11-13). How we checked →
Source and checks
This item ran in the brief of Nov 16, 2026, alongside 10 other items. How we checked it, the quote checks and what was still missing that day are on that issue's How we checked page.
Cite this item
SecondSource (2026). "A group of authors reports in their paper's abstract that adding process control outside the model, without changing the underlying model, raised the unattended completion rate on a 120-case benchmark the authors built from their own business workflows from 50% to 100%.." https://secondsource.io/en/research/2026-11-16-3/. Accessed access date.