Archives

Nous Research's new gateway mode lets solo builders ship full-stack apps from a phone. The list of supported platforms grew to twelve overnight.

Reusable, versioned agent behaviors are turning ad-hoc prompts into shared infrastructure — and the rest of the harness ecosystem is watching.

While the coding-harness crowd is fighting over benchmarks, OpenClaw is quietly attracting the indie agent builders who care about something else entirely — agents that remember.

Internal SWE-bench numbers show a 38% jump on multi-file refactors over last month's release. The trick: structured task decomposition that runs sub-agents in parallel.

Four contenders, four philosophies, two completely different problem statements. We talked to people on all four sides about what they're actually competing on — and where they're not competing at all.