Pre-1.0 · research preview · self-installed
DevPlane is a control panel for teams running AI coding agents
It assigns work, caps each run, tracks handoffs, and checks reported completion against pushed code, deployments and live pages.
The problem
When agent reports and the code disagree
An AI coding agent can say it pushed code, passed a check or deployed a change while the commit is missing, the build did not move or the live page still shows the old state. The person running the work then opens several tools, compares commits, reruns checks and reads build logs before deciding whether the task is finished. Across several projects, a capability may already exist in another repository and get built again because nobody found it. The backlog may say done while the checking and repeated work remain.
What DevPlane is
Assign, limit, verify and reuse agent work
DevPlane is a control panel for solo founders and small teams that need to give agents more work without losing control of cost, handoffs or completion.
Assign and follow work
See assignments across agents and projects, including handoffs that would otherwise disappear between repositories. For cross-repository handoffs, DevPlane can run the supplied evidence command before confirmation and store the result.
Check completion
When an agent reports a task finished, DevPlane compares that report with evidence from the repository, deployment and live site. Failed evidence keeps the work open.
Set run limits
Each run has explicit spend and memory limits before work begins.
Reuse existing work
DevPlane searches 856 declared capabilities across 30 repositories, including matches that names and filenames do not reveal.
The research
Testing risk compensation in agent work
DevPlane is also being developed as the instrument for a field study of risk compensation in human-AI coding work. The study asks whether human vigilance falls as agent reliability rises, and whether task counts miss the checking, handoffs and follow-up carried by the person in charge. Its method is designed to compare agent reports with independently established outcomes. The aim is to give teams a sounder way to judge agent work than counting completed tasks alone.
Status
Private and in daily use
Pre-1.0 · research preview
DevPlane is private, in build, in daily use and runs only on my machine. I use its completion checks, spend and memory limits, handoff tracking and cross-project search across my own work. Managed and self-operated versions are planned.
If you run several AI coding agents on real codebases and want to compare notes, send a short message.
Become a design partner →