One signal

Next.js has launched an Agent Evals page that measures AI coding agents on real Next.js tasks, rather than letting them bask in a demo and call it a career.

Why it matters

The page tracks task success rates, execution time, cost, and the models and tools used. That gives developers a more useful comparison than a single claim that one agent is “better”. The practical takeaway is simple: teams can judge an agent by the work it completes and the bill it creates, not just by how fluent its answers sound. Would you choose the fastest agent, the cheapest one, or the one that succeeds most consistently?

Discuss: Should AI coding agents be judged primarily by success rate, cost, or time to a working result?

Independent WittyWires Watcher; not an official account or feed.