Eno Reyes (@EnoReyes) on X (original) (raw)

cto

@FactoryAI

- Building software that builds software

• Pinned
  
• 
  Interested in real Open AI? Announcing Transformers-Chat, a 100% open source knowledge-grounded chatbot that allows you to ask questions and chat with the 🤗Transformers docs. Powered by Flan-UL2,
  @Langchain
  , and Instruction Embeddings.huggingface.co/spaces/hugging…1/3
• 
  Had a fascinating time at AI Ascent hosted by
  @sequoia
  . Favorite takes: ⁃
  @ClementDelangue
  : every business will have a custom language model ⁃
  @saranormous
  : teach your kids how to prompt engineer ⁃Jensen Huang: all pixels are going to be generated, not rendered
• 
  We believe that users should not need to think about managing context beyond the initial specification inside their coding agent. Droids should automatically compress context that is no longer relevant, and preserve relevant instructions and progress as it works.
• 
  How do you build a reliable agentic system? Agents have many definitions, at
  @FactoryAI
  , we consider a system “agentic” if it exhibits: Planning Decision Making Environmental Grounding 🧵 sharing highlights from our
  @aiDotEngineer
  talk about how to build reliable agents 1/5
• 
  The Factory CLI now supports mixed-model runs. You can build a specification with a SOTA model, and execute with a more cost-efficient model. Developers deserve choice. You can now get the best out of all models, from every provider. For example, Sonnet 4.5 for specification
• 
  Factory Droid power user guide for devs who want the AI to pull its own weight: 1/3 Install the IDE plugin so it sees your files, cursor, and errors while you just toss short prompts.
• 
  3 weeks ago, we were not even on this tracker. Seeing
  @AmpCode
  and
  @FactoryAI
  in the #1 and #2 spot should tell you an important insight: Software Development Agents are not about the model. They are about the product experience and the community. Really proud of the team, and
• 
  One of the things Droids do well is handle prompt caching for all providers very natively. Typical prompt caching ratio is 9:1 or greater. We've even measured that it might be better than some model providers native harnesses. Should we do a blog post on this?
• 
  Lots of customization in our recent releases. One nice QOL thing is that Droid now knows how to configure itself. Try asking droid how to add hooks to itself :)