Users can now jump in and change what an AI agent is doing while it is halfway through a task.
April 29, 2026
Original Paper
Revisable by Design: A Theory of Streaming LLM Agent Execution
arXiv · 2604.23283
The Takeaway
Standard AI agents operate as a black box where users give a command and wait for the final result. This stream paradigm allows for real-time collaboration between a human and the machine during execution. If the agent makes a mistake or heads in the wrong direction, a person can intervene and correct it without restarting the whole process. This is made possible by a new system for reversing and revising agent actions mid-stream. This shift makes AI agents feel more like a co-pilot and less like an unpredictable automated system. Real-world tasks become more efficient when the user and the agent can work together simultaneously.
From the abstract
Current LLM agents operate under an implicit but universal assumption: execution is a transaction -- the user submits a request, the agent works in isolation, and only upon completion does the dialogue resume. This forces users into a binary choice: wait for a potentially incorrect output, or interrupt and lose all progress. We reject this assumption and propose the stream paradigm, in which agent execution and user intervention are concurrent, interleaved processes sharing a bidirectional chann