I Let ChatGPT Manage My Workweek

Was this newsletter forwarded to you? Sign up to get it in your inbox.

I sat down to write my second-quarter goals at 4:30 p.m. on a Tuesday in early April. It was the day after I was supposed to turn them in when I decided to be an adult and survey the damage from the first quarter. And I do mean damage. I’d written only half of the columns I’d committed to. Another project I had promised hadn’t even gotten off the ground.

I could give the usual excuses—the quarter was busy, the project hit walls outside my control—but the real culprit was obvious: I may be a great writer, but I am garbage at project management.

For 15 years, I handled this weakness by tiptoeing around it. I didn’t take on managerial roles that would have required more organizational skills. I didn’t take on so much freelance work that I couldn’t keep the deadlines in my head. I passed on ambitious projects—too many moving parts.

This duct-taped approach worked until I decided to join Every full-time in April. If I were going to take on more responsibility as a full member of the team, I needed to get serious about project management. Which, in 2026, meant I needed to bring in AI.

So I built myself a project manager: a ChatGPT agent that holds my OKRs—objectives and key results, the goals that define a successful quarter—watches my calendar, reads my Notion to-do list, and helps me decide what to do next. Otherwise, I’d spend my day opening Slack, refreshing X, panicking lightly, repeat.

Most AI-at-work advice starts with the part of your job you’re already good at: Write faster, code faster, analyze faster, ship more. I’m interested in the other side of the equation: using AI to support the part of work that makes it hard to believe you’re good at your job.

I’ve set up project management with both my Plus One agent, Margot, and as a ChatGPT agent. I’m featuring the ChatGPT agent here, but you can create your own project manager with any system that gives you a combination of memory, context, and intelligence—more on that below.

Why AI can babysit my to-do list now

I’d tried using ChatGPT as a project manager before, during a freelance month last year when I’d overbooked myself and had deadlines staring me down like unread letters from the IRS. I would open a new chat and type some version of: “I have this deadline, this deadline, and this deadline; this meeting, this meeting, and this meeting. What should I do?”

For one-off triage, it worked well enough. The problem was the context that it had about me—or didn’t. Every time I came back, I had to explain everything again: the clients, the deadlines, the pieces in flight, the meetings, the priorities, the fact that one project was more important than another for reasons that were obvious to me and invisible to the chat window.

Then, over the past six months, several things converged to make more comprehensive project management using ChatGPT possible.

First, memory improved enough that the system could carry context and apply it across conversations. Next came advanced tool use, which enabled AI to navigate and use browsers and other tools. Integrations meant that ChatGPT could finally do things like open my Notion, check my calendar, and read my Slack. Finally, products like OpenClaw and Every’s Plus One wrapped all this firepower in a package that even I, a technical neophyte, can work with.

If you tried to do something with AI a year ago—like manage a marketing workflow or run an analysis of financial results—and it didn’t take, try again. Chances are that the model and the product around it have shifted in ways that move the finish line in your favor. It was time for me to take another swing at AI-native project management.

What I built: A project management agent

Saying “I built an agent” makes the whole thing sound more sophisticated than it is. The truth is that AI did most of the work—I just put the right information in places AI could see it, connected the tools and software where my work happens, and described the job I wanted done.

Context to shape the agent’s memory

With context, the agent can turn a vague goal into Thursday’s first task. Without it, it’s just a Magic 8 Ball for to-do lists.

So, as I was going through the setup for my agent (which you can do directly through the chat interface), I made sure to provide plenty of documentation for the agent-builder to build on top of. Most importantly, I gave it a link to a Proof document with my OKRs, four objectives, a dozen-ish key results, and a rough sense of a stack-ranking of projects. Then I asked it to do the first piece of project management I am worst at: I asked it to turn “a successful quarter” into concrete phases, milestones, deadlines, and tasks.