AI Finally Gave Me a Reason to Use Project Management Tools

My Plane Business trial expired recently. I asked Codex to check what that meant for my workspace. The free plan covered what I was using: the projects and tasks were still there, I could still read comments and view images, and nothing I needed seemed to require an upgrade.

Free is nice. The stranger part was that I had started caring about project management software at all.

Before AI, I couldn’t stand these tools.

中文版

Writing code already filled the working day. There were features to build, integrations to get working, and bugs to fix. Then came another system to maintain: descriptions, estimates, status updates, and an assortment of fields. Sometimes a missing field would stop me from moving a task to the next state.

I knew what I was working on and why. Copying that information somewhere else felt like being given another job.

Back then, keeping two or three pieces of development work moving at once was already pretty good going. Most work had to wait its turn. There wasn’t much state to remember, and I could usually keep it in my head.

The tracking systems I had used seemed to be built mainly around the needs of project managers and team leads. They needed to know who was doing what, how long it would take, and when it would ship. Fair enough. But the people doing the development had to enter and maintain all that information. Whether the system helped us with our own work often seemed to come further down the list.

I understood why a team needed visibility. From my side of the desk, though, the development work hadn’t gone away. There were just more forms to fill in.

For side projects, the case was even weaker. Maintaining one or two projects after work took enough energy. I had more ideas, but most had to wait until I had the time to build them. With so few projects actually underway, there wasn’t much to manage.

AI loosened that constraint.

Given enough ideas, working on ten or twenty small projects in parallel became plausible. One conversation could handle frontend changes while another worked on the backend. Others could investigate a problem, compare approaches, or run checks. More of my time went into explaining what I wanted, discussing tradeoffs, and reviewing the results.

That brought a different problem.

Where had this feature got to? Why had we chosen that approach? Had the code actually been changed, or was there just an implementation plan? What had shipped, and what was waiting for me to review? After leaving one conversation to deal with something else, I could come back days later and have to work out where it had stopped.

I was still one person. There was just a lot more to keep track of.

Before trying Linear, I used Apple Reminders for this.

Reminders wasn’t doing anything wrong. It was convenient for jotting down an idea or an important direction. But my entries were usually only a sentence or two. Keeping the background, implementation decisions, progress, and acceptance checks up to date took more effort. I still had to write it all myself.

So the information stayed thin while the list grew longer. The easy things got done. What remained tended to be the harder things, and looking at the list became rather discouraging. Unfinished work was enough of a nuisance without having to write more notes about it.

I started trying project management tools: first Linear, then Plane.

Two things made the difference this time. I had more work moving in parallel, so the records were useful to me. And AI could maintain those records as part of the development work.

In Codex, I can describe a feature and ask the assistant to find the existing task, create one if needed, and record the plan and implementation details. Decisions made during a discussion, changes made during development, checks that passed, and things still unfinished can all be written back to the task.

The information is already there in the conversation. I used to have to reconstruct it later and enter it myself. Now the assistant can record it while it works.

That leaves very little extra manual upkeep for me. Meanwhile, the board tells me what’s underway, what’s important, what hasn’t started, and what needs my review. I can open it and use the information to decide what to do next.

Keeping the board finally pays off.

A recent change to a web product’s navigation is a good example.

After login, the product homepage was replaced by the workspace. There was also a brief flash during the transition. I didn’t like the experience, so I raised it in Codex and asked to discuss the options before making changes.

The assistant inspected the implementation, explained the behavior, and proposed ways to separate the homepage from the workspace. Once we’d agreed on an approach, I asked it to create the tasks in Plane, document the implementation plan, and get to work.

The changes covered frontend pages, navigation, login redirects, and some supporting backend work. These became related tasks, with several agents working on different parts. Afterwards, the checks covered changes in login state, compatibility with old links, and how the pages behaved on desktop and mobile.

Anything missed during implementation was fixed along the way. Verification results and remaining work went into the task records. After I approved the release, the assistant deployed the changes, checked the live site, and updated the task statuses.

My main contributions were raising the problem, agreeing on the approach, and reviewing the result. Maintaining the record didn’t become a separate chore. Later, I could look back and see why the change had been made, what had been delivered, and what had actually been checked.

For this way of working, the AI feature I care about most is MCP support.

MCP, the Model Context Protocol, gives the coding assistant a way to work with external tools. In this case, I want Codex to be able to find tasks, update their status, and read and write comments. Discussing the work, doing it, and maintaining its records can then happen in the same conversation.

I don’t have much use for the project management platform’s own AI assistant. I’m already doing the development in Codex, where the relevant information is available. Going to another platform to ask another AI to process it again feels like an unnecessary extra stop.

I want the assistant I’m already working with to keep the tasks in order. Opening another chat just to maintain them doesn’t appeal to me.

The software’s own design and usability still matter, of course.

I liked Linear. Its design and interactions suited me, and I would seriously consider it for a company. For personal use, the price and free allowance mattered too.

At the time of writing, Linear’s free plan includes 250 issues, with unlimited archived issues. That doesn’t mean you can only create 250 issues over the lifetime of the account. Still, with several projects moving at once and a backlog of things I want to do, the allowance felt a little tight. Liking a product and deciding to pay for it are two separate decisions.

YouTrack has a fairly complete feature set, but its look and interactions aren’t quite to my taste. It now offers an official remote MCP server, so external AI tools can read and update tasks. Having a convenient, ready-made integration in my usual development environment is another part of the experience. I’d want to judge that separately from whether an integration is technically possible.

For now, I’m using Plane. The free plan covers my main needs, and the connection I already have lets Codex manage tasks. I don’t currently need the more elaborate paid features. Comments and images work for my purposes, although attachments still have limits: the Cloud free plan allows files up to 5 MB each. So far, that’s enough for my screenshots and task notes.

GitHub Projects is another option. It brings issues, pull requests, and project views together, which can make the work easier to follow alongside the code. For a project already hosted on GitHub, it’s worth considering.

I’m still adjusting how I use all this. How small should a task be? What counts as work in progress? How do we distinguish implemented code from a released feature? Which results need to wait for my review? I recently put those agreements into the project’s AGENTS.md so later AI conversations could follow them. When something doesn’t work well, I can revise the agreement.

I used to think that an independent developer using project management software was making extra work for themselves. Looking back, that judgment made sense at the time. I could only work on a few things at once, and I could remember them.

Needing it now is actually something I’m pleased about. I’ve often had more ideas than time to build them. Now I can finally try a few more.