Pinned
Factory
1,172 posts
Factory
@FactoryAI
Assembling @droid, the frontier software development agent. Available to anyone, with any model, in any IDE/terminal.
- Factory reposted
We’ve come a long way since VSCode. This is what the future of software development looks like, where you don't even view code anymore. Here’s a throwback to one of my favorite missions, now viewable on our desktop app, hundreds of file touches from a single prompt.
00:00
00:52Missions are now available on Factory Desktop. - Factory reposted
Recipe for frontier AI research: 1. Use models that support open research 2. Run a mission in the desktop app with the goal of building out the component of your pipeline you care about 3. Monitor the agents for the duration (2 hours to 2 weeks) A GUI for auto research!Missions are now available on Factory Desktop.00:00 - Replying to @FactoryAIMissions are designed to automate day-long tasks like building internal tools and performing AI research. Now, they can run anywhere you use Factory, including the CLI, Web, and Desktop App. To work asynchronously, you can also delegate execution of Missions to Droid Computers.
- Factory reposted
a new flower blooms in the model gardenClaude Fable 5, a Mythos-class model, has arrived in Droid.
- Replying to @FactoryAIMost models can look strong on constrained tasks. Fable pulls away when the work gets ambiguous, messy, and long-running. Try it today:
- Replying to @FactoryAIWe're excited to have partnered with @AnthropicAI on this launch, in early testing Fable is: • More autonomous on multi-step engineering work • Better at recovering from dead ends • Stronger at deep bug hunts and security review • Built for tasks that require hours of
- Replying to @FactoryAIEY, Adobe, NVIDIA, and Palo Alto Networks use Factory harness to build their internal software factories. The long-term default for agent infrastructure is model-agnostic. If you want early access to the Factory Router for your @droid, reach out to our team:
- Is automatic model routing slower than just sticking with your favorite model? Sessions with routing are faster on average because smaller models have lower latency per token. The routing decision itself is a single classification call at session start.
