OpenAI Codex lets users teach workflows
James Sun, a product staff member at OpenAI, announced a new capability in the Codex desktop application that enables users to teach the agent custom skills and workflows in a manner similar to instructing a human coworker. This feature supports complex workflows involving computer and browser use, which run significantly faster the second time around once the agent has observed the initial demonstration of the task.
Teaching AI agents via direct demonstration is a massive UX breakthrough that bridges the gap between static prompting and true computer-use delegation.
* Demonstration-based Learning: Moving away from raw prompting or scripting towards interactive demonstration and human-like instruction represents a major milestone in making agentic software accessible to non-technical users.
* Cached/Optimized Workflows: The significant performance improvement during the second run indicates that the agent creates optimized local playbooks or structured standard operating procedures (SOPs) based on the first demonstration.
* Browser and Computer Control: Emphasizing integration with browser and native desktop automation highlights OpenAI's strong push towards end-to-end autonomous computer-use agents.
DISCOVERED
1h ago
2026-06-19
PUBLISHED
1h ago
2026-06-19
RELEVANCE
AUTHOR
JamesZmSun