Pelicans on a bicycle and Spotify clones are fun, but I can't always wait for gpt-5.2 ultra super high to think how to charge more, composer-1 was blazing fast, which I loved. Now building new products (@automat_ai 👀), and really excited for composer-2.
Beautiful day in SF.
Today, we’re visiting @cursor_ai’s top Composer-1 users to give them a little gift.
We tracked down old-school GPUs and laser-engraved them.
Keep an eye out for a red car pulling up outside your office! We’re very grateful you’re building with us.
Our AI employees can talk now
We integrated Google Gemini 3.1 Flash TTS into @automat_ai Workforce — our digital workers deliver audio standups, briefings, and updates with natural speech.
Inline audio tags let them control tone mid-sentence. 70+ languages. Zero extra infrastructure.
Automat Workforce digital employees now have a voice thanks to Gemini 3.1 Flash TTS, so I can listen to their standup while sitting in traffic on the 101 🤖
Shot out @googledevs for giving us early access!
Introducing Gemini 3.1 Flash TTS 🗣️, our latest text to speech model with scene direction, speaker level specificity, audio tags, more natural + expressive voices, and support for 70 different languages.
Available via our new audio playground in AI Studio and in the Gemini API!
I think subagents have some use cases, but this is not it, specially with long context and compaction, I don't want Opus to hand critical tasks to subagents. I don't care if it's Composer 2 or Haiku.
@pirchavez We are not. Opus can choose to spawn subagents, which default to a faster/cheaper model.
This is no different than Claude Code, where the Explore subagent uses Haiku, not Opus.
You can define custom subagents which use Opus, if you prefer.
Announcing Codex.
A new product from OpenAI that moves beyond coding, into cooking. We were already cooking before, but now *you* can cook too ... with Codex. It is powered by the same technology as our other Codex products. You can just cook things.