Windsurf: Devin Bots Own an Outcome and Run Sessions Until It Is Done
Windsurf's Devin Desktop introduced Devin Bots. A user gives a Bot an outcome to own in its chat, and the Bot plans the work and runs sessions until the work is done. The release also added zoom buttons and keyboard zoom shortcuts to the image viewer.
Key Takeaways
- Devin Bots own an outcome, so users describe the end goal in the Bot's chat instead of managing each step.
- Each Bot plans the work itself and decides what sessions are needed to get there.
- A Bot keeps running sessions until the work is done, which reduces the need for follow-up prompting.
- The feature moves Devin Desktop further toward managing fleets of agents rather than single chats.
- The image viewer gains zoom buttons and keyboard shortcuts, useful for reviewing visual output.
- Devin Local now retries stalled model streams after 40 seconds and keeps images beyond 20 per conversation, making long runs more dependable.
Devin Bots Arrive in Devin Desktop
Windsurf's Devin Desktop release v3.14.30819 introduced Devin Bots, a new way to hand off work. Instead of prompting a single session step by step, a user states an outcome in the Bot's chat. The Bot owns that outcome from then on.
How a Bot Works
Once a Bot has an outcome, it plans the work needed to reach it. It then runs sessions on its own and keeps going until the work is done. This shifts the user's role from steering each session to defining the goal and checking the result. For teams already running several agents from the Agent Command Center, a Bot adds a layer that coordinates sessions toward one objective rather than leaving that coordination to a person.
Other Improvements in the Release
The image viewer now includes zoom buttons and supports keyboard zoom shortcuts, which helps when inspecting screenshots and design output from agent sessions in detail.
Windsurf also shipped a range of fixes in the same release. Repeated git push commands to the same branch and commit now collapse into one worklog row. Devin Local keeps images in a conversation until the model provider's request limit is reached instead of dropping them after 20. Devin Local also retries a stalled model stream after 40 seconds instead of waiting about 4 minutes, and long Devin Local sessions on Anthropic models keep the prompt cache across turns.