Project Management Lite With AI
In my experience, sooner or later every project needs some sort of management.
If you’re lucky, you can get by with a TODO.txt or maybe TODO: comments in
your code. If you’re really lucky, someone else does the project management. But
what if you need more than a text file but nothing approaching Jira
? That’s
what I’ve been sorting out for the last year or so.
This acts as an unofficial follow-up to my How I Plan Work with Agents . Read that for what I do instead of how I do.
The unbearable weight of Jira
If you’ve never been involved in project management before, it can be a lot. Not only is there, you know, work that needs doing, but now you have to figure out how to track the work that’s waiting or in progress or done or was a dead end. This snowballs until you’re dealing with company and team priorities, team vacations, some sort of work-tracking board, grooming, sprint planning, and on and on. If it has a ceremony attached to it, you probably have to do it 1.
I’m just one person, though, working on a project of my own. One human plus AI agents just doesn’t need a Jira board 2. I already showed you how I plan work with agents so this is more of a tool guide than a how-to. This has been my go-to approach while working with AI agents. So, I’m assuming you’re like me and using agents. You’re also tired of trying to keep track of which agent is doing which task and what still needs doing.
Let there be lite
In programming, a lot of people follow a rule of proximity: code pieces that
work together should be near each other in the code base. This can’t always be
the golden truth, of course, but you do your best. If that’s true for a
function, it should also be true for the rest of the project. The docs/ folder
lives in the repo because it belongs to the project, for example. I’m a firm
believer, too, that everything in your project should be readable by you, if
you use agents or not.
I don’t want to pay for a service like Jira. I don’t want to maintain another server somewhere to run a self-hosted service. I don’t want to use GitHub’s or some other forge service’s planning tools. If I do that, now my tasks live on GitHub and not my local computer. I shouldn’t have to have Internet access just to see what I need to do next. I don’t want something that would be hard for a contributor to use. I want to write code, not sort out ticket tracker issues.
Basically, project management needs to be files that my agents and I can read and update easily.
beads
Currently, my stack consists of two pieces, all free and open to use. First, we
have beads
, which is my issue tracker of choice. It lives on the command
line and your agents can use it easily. There are skills available for most
harnesses and quite a few different
TUI and
GUI applications for human use. As
I mentioned in my previous post
, I usually have an agent break apart my plans
and save them as beads.
Using beads, I can prioritize tickets one over the other, I can mark them as
dependent on each other, I can label them by domain, and the app provides a
ready queue of work that’s not blocked or claimed. Using these features, agents
can work, often in parallel, without stepping on toes or blocking each other’s
work. And with how beads syncs via your git repository, there’s nothing to
keep in sync between users. Do a git fetch and you’ll have the new beads.
With bigpowers comes big docs directories
Yeah, this one’s a little trickier. If you aren’t a little careful about it, this set of skills can swamp you in Markdown and YAML files. Most of them are good to have around, though, so I don’t complain much 3.
Before I start agents on a feature or a fix, I need to understand the work
myself. This is where we get a little closer to project management and not just
keeping track of tickets. bigpowers
gives us a more visible paper trail of
work and planning than beads alone, since full documents usually care more
information than a single ticket. That last bit’s important: you need to know
how and why to do something just as much as you need to know what
that something is 4.
bigpowers shines, for me at least, in planning. It’s a good tool for keeping
scope creep low and for justifying decisions with
ADRs. If you aren’t familiar
with ADRs, they’re documents used to describe what you need and why, which
options you’ve tried, and what you’ve settled on…and why. They often come out
of spike/research tickets but don’t have to. bigpowers helps you create and
keep specs, plans, and the like up-to-date in Markdown files, an easy format for
humans 5.
Do bigpowers and beads overlap? Yes, quite a bit in some ways. They’re both
for tracking work. You could get by with just bigpowers but I find beads to
be more intuitive for how my brain works.
What about the agents?
You’ll need to install a different extension or set of skills for either of
these tools if you want agents to work with them. Since I’m currently using
Pi
, I have the pi-beads
extension installed. If you don’t see skills for
your chosen harness or agent, I’m sure your agent could convert existing ones.
Actually using this
I’ve been following some variation of this approach for a year or more. It has been extremely helpful for both work and home projects.
Good
Since the trackers are machine-readable and -updatable, agents can claim, update, and close tickets all on their own; I don’t have to keep track or translate anything. But, if something is wrong, I can correct it by either editing the bead or the related Markdown planning documents. And since agents only act on ready work, you can concentrate on getting more work ready without an agent trying to preempt you.
Bad
But, adopting this approach can have some downsides depending on your
environment. If you’re trying to use this for tracking cross-team initiatives,
it’s going to be clunky. If you need emails or permissions or, I dunno, Gantt
charts, this isn’t the solution you’re looking for. Neither of these tools make
it easy for searching across multiple projects although the Markdown files are
extremely searchable. If you’re working with a lot of non-technical people, this
might be a more difficult situation since they’ll have to read Markdown files or
you’ll have to set up some kind of beads GUI for them.
These aren’t the greatest roadblocks but it’s important to know about them. If a system doesn’t work for you, don’t use it. Find the tools that make you able to do the work how you think, not someone else.
Ugly
It can take a bit of setup, even just for yourself. That’s a one-time operation, though, so I think you’ll make it. If you’re like me, though, you’ll be tinkering for days.
Once a project’s team grows beyond a couple of people and agents, someone is
going to ask you for a dashboard of your progress, audit trails for compliance,
or some other business thing that these tools, especially beads, don’t make
easily available.
If you find yourself needing project management for 40 people, you’re going to want to find something else. Luckily, that path has been under development for a long time. You’ll find something.
Verdict?
I plan to stick with this approach, at least for personal projects. I’m sure
I’ll evangelize beads at future companies, along with other winning tools. If
you try it out, let me know how it goes 6!
I’m sorry you have to do all of that. I’ll stick to code, thank you. ↩︎
How do you hold standup with software anyway? ↩︎
Not too much anyway. ↩︎
When is always yesterday, who is inevitably you, and where is none of my business. Work where you’re comfortable. ↩︎
I find it’s best to not think about the YAML. ↩︎
I never know how to wrap these up. ↩︎