Launching today
Planwright
The Control Plane for Agent Labor
13 followers
The Control Plane for Agent Labor
13 followers
Post-kanban, agent-native planning and control. Humans write objectives. Claude Code, Cursor, and Codex decompose, execute, and check in. Every decision signed. Every change audited. Compose your Agentic Engineering stack with PlanWright and we'll take care of the accounting for the auditors.
This year I opened a venture studio where we went really fast launching new companies and building products. One of the things we learned was we need to get way more organized around how we plan products and give better tools to our coding agents in order for them to be the most effective, and to ensure we retain human control and review.
At the same time I happened to be supporting an initial SOC 2 audit and freaked out about how these new coding agents will never pass, especially as we start to move toward more agent autonomy. So I decided to go down the rabbit hole of investigating what tools exist on the market to allow for agentic engineering to happen and at the same time pass audits.
This product is the result of that research and the new audit standards that came into being at the beginning of 2026 for Agentic Engineering flows where human still need to define objectives and control acceptance of coding efforts. However, we still want coding agents to contribute code at a rapid pace and report progress and breakdown tasks. Moving toward a Dark Factory coding model we are going to need robust planning and control mechanisms.
When we wrote this tool and started dogfooding it in our own projects, what we found out was this is way more capable as an enterprise planning tool to retain complex context through projects and we have actually extended the capabilities of our PMs to be able to synthesize Objectives from chaotic inputs and deliver specs that are machine ready by living inside of claude desktop.
How does Planwright actually verify what the agent did in the IDE matches the objective it was given, especially if the work happens across multiple sessions or branches?
@zilan42dm the MCP connection gives instructions to your coding agent on. how to report and what actions to take in a well behaved Engineering SDLC motion. One Agent instance can claim any objective and can part of that objective contains Acceptance Criteria that must be met for completeness. The coding agent is responsible for implementing those criteria and is also required to report on test runs. The ultimate check is after the coding run in complete, the Objective is passed into the "Request Acceptance" lane for humans / human+agent checks the work, including completeness.
How does the "every decision signed" part actually work in practice, like is there a way to roll back or contest an action an agent took autonomously?
@glerioqv the signing part is for the new audit standards to ensure that code commits correspond to Objectives written by a human. If something goes wrong where you need to back out a commit / PR that's fine, because it's still just github underneath. This would take place at a Human review gate in 'Acceptance.'
The objective would typically not be Accepted and would be rejected for not meeting specific acceptance criteria by the human/human + agent. this places the objective back in the queue in Scheduled or Backlog where another agent can pick it up later for rework.
How does Planwright handle conflicts when multiple agents (Claude Code, Cursor, Codex) are working on overlapping objectives at the same time, and is there a way to manually intervene if an agent goes off-track?
@bernahru8 each objective can only be claimed by one agent at a time. ('In Progress'). We ensure the overall plan is consistent by performing quality checks when initial Objectives are injected on to the board. If those new objectives overlap or are mal-formed, planwright will reject them and/or mark them as unclear/needs refinement for you.
Tried it with a Claude Code session and the audit trail actually showed every step the agent took, which I didn't expect to be so clear. The sign-on-every-decision thing feels a little heavy at first, but it makes handoff to the next agent painless.
@mervetc6u we wouldn't have naturally wrote the sign-on-every-decision, but there are audit requirements in Enterprise environments where this is necessary if we want to let these agents go on their own. That's the only way to go fast and we will ultimately need an audit trail.
Finally tried this with our Cursor setup and the audit trail alone is worth it. Love that every decision gets signed automatically, no extra work from the team.
@erafettinkdku2 no TPS reports!
love how planwright keeps a clean paper trail between objectives and what the agents actually ship
@balcicek59166 amen brother! it's all the paperwork we love to see and hate to write.