I turned ChatGPT, Claude, and Gemini into project management apps, and one clear winner emerged

Large language models (LLMs) might be the most accessible and flexible technology we've ever had access to. By default, they're presented as chatbots—but that's just one way to interact with them. With the right prompts, you can transform an LLM into something that behaves like an entirely different app, complete with its own custom interface.

I've been doing exactly that by prompting LLMs to behave like project management systems. I turned ChatGPT, Claude, and Gemini into project management systems to see how well each one handled the job. Here's how I did it—and which one came out on top.

ChatGPT as a project management system

This is where it all started

Back in 2024, I was using ChatGPT heavily—repurposing it to function like different tools rather than just a chatbot. One of my most successful experiments was turning ChatGPT into a project management system, and a lot of you really seemed to like it.

At the time, ChatGPT had just introduced its Memory feature, which meant it could retain information across chats. I combined that with a carefully crafted system prompt to create what was essentially a custom project management mode. I could tell it to save tasks to its memory and remove them once they were complete. Better yet, I could send a simple message like "Start PMS," and it would retrieve every task stored in its memory and present them in a structured list.

Not long afterward, ChatGPT introduced scheduled tasks, and I repurposed them as a reminder system. Every morning, ChatGPT would send me a summary of everything I still had left to do. It worked remarkably well, and felt pretty futuristic.

As a personal project management system, the overall setup worked surprisingly well, but something still felt off. The biggest limitation was that everything lived inside a chat interface—my tasks existed only as text in a conversation. I tried displaying them in tables, but it still didn't feel like using a real project management app. I couldn't quite solve that problem until Claude came along.

Claude as a project management system

Access to other apps and the file system changed everything

When Anthropic released Claude Cowork, it was a complete game changer. The LLM now had direct access to my file system, which meant that instead of relying on its memory feature, I could save all my tasks to an actual file on my computer and have Claude interact with it directly—or edit it manually myself.

Cowork also introduced a plugin ecosystem, and one of the official plugins they bundled in was called Productivity. Using it felt like Anthropic read my ChatGPT PM article and created an upgraded version.

First, it creates a TASKS.md file on your computer where all your tasks live. It also includes an HTML dashboard template. Feed your TASKS.md file into it, and your tasks are automatically displayed on a Kanban board, making them much easier to visualize and manage—effectively solving the biggest limitation of the project management system I had built with ChatGPT.

From there, you can create new tasks directly from the chat interface, view everything on the dashboard, and, because TASKS.md is just a plain text file, open and edit it with any text editor whenever you want.

However, that's only part of the story. The Productivity plugin also includes a collection of skills that streamline task management. One of them is called Update. If you connect Claude to apps like Asana, Notion, and others, then run the Update skill, it retrieves the tasks scattered across those apps and consolidates them into your centralized TASKS.md file.

This workflow turned out to be genuinely game-changing for me. Even today, Claude serves as my central project management hub—the single layer I use to orchestrate all my other productivity apps.

Gemini as a project management system

It’s too isolated—unless you go Antigravity

After ChatGPT and Claude, I figured, why not see if the other LLMs could do it too? Gemini was the obvious next choice, and on paper, it seemed like it should work perfectly. Being a Google product, it has access to Google's entire ecosystem—Google Tasks, Google Keep, Gmail, and more.

Technically, you can orchestrate a complete project management system around that. Gemini can reference your emails, captured thoughts, and NotebookLM notebooks, synthesize everything, extract actionable ideas into Google Keep, and turn your to-dos into Google Tasks. For collaboration, it can organize those ideas into a Google Doc that you can then share with other people.

That orchestration does work, but it felt disjointed and nowhere near as streamlined as I wanted it to be—at least in the web app. However, the mobile app—especially on Android—offers a much better experience. What really stood out to me was how effortless it makes idea capture. I can snap a photo, record a quick voice note, or share a screenshot of something I'm reading online, and Gemini can process that content and seamlessly turn it into tasks, notes, and even Calendar events.

As an overall package, though, it still falls short of what I get with Claude and ChatGPT Work. The biggest limitation is that Gemini doesn't natively integrate with third-party project management platforms like Asana or Notion—a clear advantage that ChatGPT and Claude have. The main workaround is Antigravity, Google's desktop application. Like Claude's and ChatGPT's desktop apps, it gives Gemini access to your local file system and lets you connect MCP (Model Context Protocol) servers for Notion, Asana, and other services, allowing you to recreate much of the same workflow.

LLMs and MCPs are making tech unimaginably more flexible

Out of the three, my favorite setup is still the one Claude made possible. But when you break it down, the entire workflow is powered by just two things: custom prompts and MCP servers connecting the LLM to my other applications. As more AI chatbots adopt MCP, the same workflow becomes increasingly portable. As you've seen with ChatGPT Work and Google's Antigravity, you don't have to use Claude to build it.

That's the real beauty of LLMs and MCP. They let you use plain natural language to shape software into tools that fit your workflow in ways that simply weren't possible before—and the creative potential is enormous. I've explored that idea by turning LLMs into project management systems, but with the right prompts and a connection to your source of truth, you could just as easily make them behave like almost any other application.