Thoughtworks Radar tells teams to size review capacity before adding agent sessions

Under the theme "Putting more agents to work," the board describes developers running several agent sessions in parallel and steering each one, and handing agents larger jobs such as unattended overnight builds that developers review afterward. Thoughtworkers mostly run multiple sessions under direct control, which the board calls "a sensible place for engineers to start."

The board warns that when review can't keep up, oversight risks becoming "human-in-the-loop theater," and that changes can arrive faster than a delivery pipeline can handle. It suggests path-to-production mapping to find where work queues up, and says teams should consider how much their review and delivery processes can handle before adding sessions.

Two other themes bear on the same work. On harness engineering, the board calls the harness "a living asset" that needs ongoing ownership, and points to an emerging plugin ecosystem built around Agent Plugins, agent plugin marketplaces and Agent Package Manager. Under "Beyond frontier models," it says the gap between open-weight and proprietary frontier models is much smaller than twelve months ago for many tasks, names the GLM family as a credible alternative for some coding-agent workloads, and cautions against blanket frontier model use in favor of model routing.

These are the board's judgments, drawn from its members' project experience, and the themes page gives no measurements. We read the themes page. We did not read the individual blip write-ups or the PDF.

Why it matters: If you are adding agent sessions, measure what your review and delivery steps can absorb first. The Radar names review capacity as the limit but gives no figure for it. Test the open-weight claim on a sample of your own tasks before it changes a model contract, because the Radar reports a judgment and not a benchmark.