You know the feeling. You crack open Jira, Linear, or whatever monolithic tool your team swore by, and you spend the next hour configuring workflows that still don’t match how you actually think. The board is too rigid. The automation is a joke. And when you try to get AI agents involved, forget it—the vendor’s roadmap is six quarters behind your reality.
So you do what any self-respecting builder does: you start sketching out your own tool. And then you realize something dangerous. Building your own project management system isn’t about tracking tasks. It’s about taking back control of your workflow—and in the age of AI agents, that control is political.
I spent the last few weeks building a project management tool on Cloudflare Workers. Not because I need another side project. Because I wanted to see if I could escape the tyranny of off-the-shelf software and assemble my own orchestration layer from composable primitives. And here’s what I found: the tool itself is trivial. The real lesson is about who gets to run the fleets.
The moment you build your own tool, you stop being a consumer of infrastructure and start being its architect.
Let’s start with the itch. You’ve probably had it. The generic PM tool that makes you fill in custom fields no one reads. The comment threads that spiral into Byzantine debates. The dashboard that shows you thirty metrics you don’t care about. It’s not just friction—it’s a slow erosion of your agency. You’re not using the tool; the tool is using you.
So you build. On Cloudflare, you get Workers for compute, Durable Objects for state, Queues for async work, and a global edge that makes your stuff fast everywhere. You stitch together a few endpoints, a simple UI, and suddenly you have a system that does exactly what you need. No fluff. No enterprise sales pitch. Just your logic, your rules, your data.
But here’s the twist that most people miss: how you ship this tool matters more than what it does. If you release it as a binary, you’re giving anyone the power to run their own control plane. If you release it as a SaaS, you’re centralizing that power in your servers. The distribution model is the real architecture.
Think about it. In an AI-native world, the ability to coordinate fleets of agents—scheduling, assigning, resolving, retrying—is the new bottleneck. The company that controls that coordination controls the value chain. If you build a tool that runs locally or on your own edge, you’re democratizing that control. If you force it through a cloud SaaS, you’re just adding another layer of dependency.
I’m not saying you should never use a vendor. But I am saying that the default assumption—that you should buy a PM tool and adapt to it—is backwards. You should build the tool that adapts to you, and then decide who gets to run it.
This is the uncomfortable side of building on Cloudflare. You’re escaping one lock-in (Jira) and embracing another (Cloudflare’s platform). But that’s not hypocrisy—it’s strategic tradeoff. Cloudflare gives you the primitives to compose your own logic, and you can still export your data, run your code elsewhere, and maintain ownership of your workflows. The constraint is platform-level, not product-level. That’s a difference worth paying for.
Real talk: my tool is ugly. It’s a few endpoints, a JSON store, and a webhook that pings my Discord. But it does exactly what I need. It tracks issues, assigns them to agents, retries failures, and surfaces bottlenecks. And because I built it, I can change it in five minutes when my process changes. That’s the kind of freedom no vendor can give you.
So if you’re sitting there staring at your backlog, feeling the mismatch between your workflow and your tool, stop waiting for the next SaaS update. Open a terminal, create a Worker, and start assembling your own control plane. The infrastructure is ready. The only thing missing is your decision.
The best project management tool is not the one you buy. It’s the one you build—and the one you control.
Now, about distribution: if you build something good, don’t just launch a SaaS. Consider shipping it as a binary or a Docker image. Let people run it on their own stack. That’s how you build a movement, not a user base. And in the AI age, movements are the only thing that matters.
FAQ
Q: Isn't building your own PM tool a waste of time when mature tools exist?
A: Yes, if you just need to track tasks. But if you're orchestrating AI agents, the off-the-shelf tools are a year behind. The real value is building a control plane that fits your exact workflow and can evolve with you.
Q: What's the practical implication of the distribution argument?
A: If you ship a binary, anyone can run the control plane on their own hardware, decentralizing coordination power. If you ship SaaS, you centralize it. In an AI-native world, that choice determines who governs autonomous work—you or a vendor.
Q: Isn't building on Cloudflare just swapping one lock-in for another?
A: It's a different kind of lock-in. Cloudflare locks you into its runtime, but you retain full control over your data and logic. You can export and migrate. With Jira or Linear, you're locked into their product decisions. The tradeoff is worth it when you need composability.