Anthropic had an agent edit its own launch video: 17 takes, 25 GB, no video editor opened. The surprising part is that the edit was text. I compare four ways agents hold an edit, list 2026 tools, and build a small editing agent you can run.
I build Tokenade, which sits between the agent and the model, so the cost of a text edit is one I can see: how much of the edit state gets re-sent on each call. A long edit session pays for every earlier take each time it calls the model, if the whole list stays in context.
For further actions, you may consider blocking this person and/or reporting abuse
We're a place where coders share, stay up-to-date and grow their careers.
Top comments (1)
I build Tokenade, which sits between the agent and the model, so the cost of a text edit is one I can see: how much of the edit state gets re-sent on each call. A long edit session pays for every earlier take each time it calls the model, if the whole list stays in context.