
I spend most of my day working with coding agents, and every day I struggle to keep track of what the AI did. Traditional project management workflows only account for work done by people. Yes, we do have humans in the loop to monitor what we are shipping, but it still takes a lot of effort to keep up with the pace. This started as a small thing I did for myself, not a product decision, but one stage we added to the workflow made things much simpler for our team.
By evening there would be pull requests open that I had asked for at some point, and I could not tell which one was which. I knew roughly what I had handed over. I did not know what had come back, what I had already looked at, or what was still waiting for me.
So I added three stages to my own board, right after Todo: Agent Todo, Agent Working and Agent done.
It is not a big idea and it took about a minute. But I stopped losing things, and it kept being useful, so this afternoon we shipped it: Agent Todo is now part of the default workflow for new Roomie projects, the same way Todo is. Existing projects keep the workflow they already have. We did not want to rearrange a board someone is already working on.

The Real Problem Is Not Organisation
It is easy to read this as "stay organised while using AI", but that is not really what happened.
For most of my career, the slow part of building software was writing it. Review was the quick part at the end, a few minutes after hours of work.
Agents flipped that. Writing a change is now fast and cheap. Reviewing it is not, because review still happens at human speed and I still need to understand the whole change.
The slow part moved from writing code to reviewing it.
Once I saw it that way, the pile of pull requests was not a tidiness problem. It was a queue of work waiting for review, growing faster than I could clear it. And nothing on my board showed it.
Why a Normal Board Hides It
Every board I have used assumes a person did the work.
Todo means nobody has started. In Progress means someone is working on it. Done means it is finished and shipped. After In Progress, the next step is "ship it", because the person who built it checked it along the way.
When an agent does the work, there is a step in between that none of these stages cover: the work is finished, but nobody has looked at it yet.
There is no column for that. So the task either stays in In Progress, which is wrong because nothing is happening, or moves to Done, which is worse because nobody has reviewed it. Most of the time it does neither and just sits in the pull request list.
That was the real issue. The review queue was there anyway. The board just did not show it, so the only place it lived was in my head, and after a long day of switching between tasks my memory is not reliable.
What the Three Stages Do
Agent Todo is work I have decided to give to an agent but have not started yet. Before this stage, I kept "I should get the agent to do that" in my head, which is exactly where things got lost.
Agent Working is work that has been handed off and is running. When something is here, I know not to start it myself, and I know a result is coming that I will need to check.
Agent done is the one that matters most. The agent has finished, there is a pull request, and no person has reviewed it yet. A normal board does not have this stage, and it is the queue this whole problem was about. The other two stages are there to support it.
After that, the normal review and done stages work as they always did.
The most useful part is that the number in Agent done tells you something. The screenshot above shows ten. That is ten finished pieces of work waiting for my review. Before, I could not see that number, so I would happily hand off four more tasks and only find the backlog late at night.
What Changed Day to Day
The main change is simple: in the morning I open the board instead of the pull request list.
The pull request list is sorted by time and does not tell you what you were trying to do. A board column does. Weeks later, the task still says why it was created, which a branch name never does.
The second change surprised me. I got better at stopping for the day. When I can see the review queue and it is long, "I will look at this tomorrow" feels like a decision. When I could not see it, every unreviewed pull request felt like something I was quietly getting wrong.
Why We Shipped It as a Default
We could have kept this as a template that people add if they want. Boards already let you add any stage, so technically nothing new was needed.
We made it a default because the people who need it most are the ones who will not think to add it.
If you already run agents every day, you have probably felt this problem and would build your own version, like I did. The person we want to help is someone who starts using agents next month and does not yet know that review is about to become their bottleneck. Without this, they would find out the same way I did, after losing track of things for a few weeks.
Our view is simple: if your team is going to run agents, the work they produce needs a place to wait that is not someone's memory.
This only applies to new projects. If you already have a board you like, nothing changes.
I Do Not Know If This Is Right
To be clear about what this is: it is one person's workflow that worked well enough on my board that we made it a default the same day. It is not a methodology, and I have not compared it with other approaches.
There are a few things I am not sure about:
- Three stages might be one too many. Agent done is clearly worth it. Agent Todo is useful for me because I hand off work in batches, but if you start tasks the moment you think of them, you may not need it.
- It may not work the same for a team. Everything here is about one person's attention. I have not yet seen what happens when four people are handing work to agents on the same board.
- The names might be wrong. The stages are named after what the agent is doing. "Awaiting review" would describe what I actually care about, which is what I still need to do. I am not sure which is better.
If you run agents every day, I would like to know how you handle this. I have not seen many people talk about it. Either it is not a common problem yet, or it is and everyone is solving it on their own.
I think it is the second one.
If You Want to Try This in Roomie
New projects already come with the agent stages. If you want to change them or add your own, workflows are on the project page, not in global settings. Open Projects, click your project, scroll past the member list to Workflows, and click Manage.

State flow is the list of stages, in order. The first and last are the start and end, and everything in between you can name and reorder. Click Edit in that section to add a stage, rename one, or drag it to a new place. Our default sits in the middle: Todo, then the agent stages, then the normal review and done stages.
A project can have more than one workflow. The list on the left shows each one with its number of states, and one is marked Default - that is the one new tasks follow.
Custom fields are on the same page, just below the state flow.

These are extra fields on every task in that workflow: a client code, a severity, a QA owner, or any dropdown your team keeps rebuilding in a spreadsheet. Click Edit, add a field, pick its type, and for a dropdown add its options. They then show up on each task, and you can filter the board by them.
Two things to know before you change anything:
- Changing a workflow changes it for every task that uses it. If you remove a stage, the tasks in it have to move somewhere, so plan that first.
- Creating a new workflow is often safer than editing the default. If you want to try something different on one board, create a second workflow and use it for that project, instead of changing the one everything else depends on.



