Working in Parallel
Most agent tools run one agent at a time, and for a reason: two processes editing the same checkout will corrupt each other’s work, so parallelism is either forbidden or left as your problem. Fletch is built the other way around — running many agents at once on the same repository is the default mode, and the architecture makes it safe rather than asking you to be careful.
Collisions are impossible, not just unlikely
Section titled “Collisions are impossible, not just unlikely”When you spawn an agent, Fletch gives it its own complete checkout of the repo — a
shared clone at ~/.fletch/workspaces/<id>/, started from
a pinned fork-point commit on your base branch, with its own .git and eventually its
own branch. Two agents “editing the same file” therefore isn’t a race that locking
prevents; the situation cannot arise, because each agent’s file is a different file in a
different checkout. There is no shared working tree, no lock to contend on, and no
ordering to get wrong.
The sandbox closes the remaining gap: an agent physically can’t write outside its own workspace, so it can’t reach into a sibling’s checkout either. Divergence between agents becomes an ordinary git problem — branches that merge at PR time — instead of a live filesystem hazard.
And because a clone borrows your repo’s history instead of copying it, spawning agent number ten costs the same as agent number one: kilobytes and milliseconds. The economics are what make parallelism practical, not just possible.
What that unblocks
Section titled “What that unblocks”Dispatch several tasks at once. Break work into independent units and hand each to its own agent. Five tasks progress simultaneously instead of queueing behind one assistant.
Race approaches. Give two or three agents the same prompt — or the same task with different agents or models — then keep the best result and discard the rest. A discarded attempt costs a directory delete.
Keep throughput while you review. While one agent is thinking, review another’s diff, ship a third’s PR, and start a fourth. You’re never blocked waiting on a single agent’s turn to finish.
You preside; the agents work
Section titled “You preside; the agents work”Running a fleet changes your role from typing to directing. You spend your time framing tasks, reviewing diffs, and deciding what ships. Fletch’s job is to make that oversight cheap: a normalized chat view of every agent, live diffs as edits land, and one Git panel to commit, push, and open PRs — plus workflows when the pattern is repeatable enough to automate the supervision too.