The Software Factory: How a Swarm of Open Models Can Out-Code the Frontier

Conventional wisdom says a software factory should run on frontier models - which are also the most expensive and typically closed-source. This talk challenges that. Coding is uniquely suited to parallelism: tasks decompose, and every attempt is verifiable against tests, so you can generate many candidates and keep only the ones that pass. Combined with the cost and scale advantages of open models - running a swarm of them in parallel can significantly outperform a single closed frontier model.

We ground this in two case studies: an enterprise maximising performance on long-running coding tasks, and a fast-growing AI-native company building code-review agents. You'll leave knowing how to architect a software factory with open models, and the results and bottlenecks to expect when you do.