A new leak suggests OpenAI is building a powerful agent-building feature for ChatGPT, codenamed "Hermes." The scoop comes from Tibor Blaho, a well-known AI product watcher, who posted screenshots showing a new "studio" interface for creating automated assistants.
https://twitter.com/btibor91/status/2046545878538961304
The screenshots reveal a builder that lets you create agents from templates, schedule them to run automatically, and equip them with connections to Slack, various apps, custom skills, files, memory, and specific instructions. The feature's landing page at chatgpt.com/agents carries the tagline "Keep work moving 24/7 with agents."
Templates spotted in the interface include "Team chat Q&A," "Morning planner," and "Bug triage." The editor page shows buttons for creating, scheduling, and previewing agents, alongside options to browse apps, add skills, upload files, and configure memory.
OpenAI hasn't officially announced the feature yet, and trying to access the builder directly at chatgpt.com/agents/studio/new currently gets blocked by a Cloudflare challenge. However, the page's source code is already pre-loaded with frontend resource names like builder, templates-modal, slack-handle-view, and skills-modal—all of which match the modules seen in the leaked screenshots. This move into persistent, scheduled agents follows OpenAI's recent major upgrade to its Agents SDK for building persistent AI assistants, signaling a clear push to make AI more proactive.