A workflow is a structured sequence of agent and human steps, defined once and run every time, with a clear trigger, a defined goal, and a complete record of what happened and why.
Every developer prompts, reviews, and approves AI output differently. What gets checked depends on who's checking, with little standardization across teams.
A change gets approved in a comment or a nod across the room. There's a trace, but nothing structured, and no easy way to find it later.
Fixing one broken review habit means retraining every individual engineer, one at a time, forever.
Upsun Dispatch replaces ad hoc habits with a workflow defined once: what an agent does, what it has access to, where a human decides, and what happens next.
Workflows remove the variance between individual habits and replace it with one defined process the whole team shares.
A trigger ties the workflow to an event in a tool the team already uses: a pull request opens, a commit lands, an issue is assigned to Dispatch or Dispatch is mentioned, a scheduled time hits, or someone starts it manually.
The agent runs in an isolated container with a fresh copy of the repository, then does the defined work: reviewing a change, posting comments, or deciding whether a push needs a fresh review.
The workflow pauses at any step built to support approval. Nothing moves forward until someone on your team says yes.
The workflow completes, and everything it did, every step, every decision, is logged against that run.
Approval happens within your Dispatch team before anything ships, wherever the workflow calls for a human decision.
Run dependency scans and routine fixes, with a human only pulled in when something needs a decision.
A workflow can span an issue tracker and a repo, without needing everyone to work inside one dashboard.
Every workflow run is a record: what happened, who approved it, and why, ready before anyone has to ask for it.