OpenAI has added Computer History to ChatGPT. This feature lets ChatGPT learn from everything you do on your computer. It can now suggest new skills and automations based on your actions. Users can review all their Computer History in a timeline view.
OpenAI also launched the Decisions API. This tool lets apps use real-time decision-making, powered by GPT-6 Luna. Developers can define questions and possible answers for agents to classify content, route requests, or pick the next action. The Decisions API is already used for support classification and internal workflows at OpenAI.
Agents now use screenshots, accessibility trees, the Document Object Model (DOM), Playwright, and generated code. This gives them more context and helps them debug and recover from failures. Agents can now complete some tasks faster than the average human.
The new stack includes async tool calling, mid-turn steering, WebSockets, UltraFast inference, prompt caching, and server-side compaction. These features make agents more responsive and lower the time it takes to get results.
Why it matters
Teams can automate more tasks, improve testing, and rely on agents for faster and more accurate results.
What to do
Try Computer History in ChatGPT to improve task suggestions and automations. Use the Decisions API to add real-time decision-making to your apps.