Where workflows benefit from flexibility
Workflows are useful because they make a complicated process legible: give an agent a task, run a series of steps, return a result. Repeatable work deserves repeatable patterns.
But engineering work is also full of useful interruptions. A test fails for an unrelated reason. A reviewer wants a diagram. The best next step is to inspect production, wait for CI, ask a question, or split the work. When those moments cannot be expressed with the tools available, the workflow becomes a constraint. The workaround is usually a bigger prompt, more hidden state, or an even larger orchestrator.
Capabilities compound
A capability is a durable thing that can be used in many situations: creating an isolated worktree, sharing a screenshot, leaving a message, running a terminal session, waking an agent later. None of these decides what the work should be. Each makes a larger set of good decisions possible.
Dispatch explores what happens when small, inspectable primitives are available alongside workflows. They can support repeatable patterns where that helps, or adapt when the work needs a different shape.
The same workspace for people and agents
Shared state, concrete tools, and an honest record of what happened can make handoffs clearer for people and agents alike. A terminal can continue; a worktree can remain available; a screenshot, pin, diagram, review, or Brain entry can be found when the next person arrives.
Dispatch explores how practical it is to collaborate with AI like another engineer—one that can keep working, leave evidence behind, and be met again in the same place. It does not ask people to surrender judgment or context to an orchestrator.