I’ve been working with AI long enough to be past the “please write an epic poem about my dog” phase and into something that actually moves my work. The real shift for me, in the last couple of months, has been using tools like Claude Cowork and Cursor. It’s not because they generate better prose than the chat window did, but because I can finally hand them tasks and not just questions or statements. That could be prepping for a sprint review by synthesizing two weeks of stand-ups and cadence meeting notes. Or it could mean triaging a product backlog with two years of Jira history. Or, pulling a project update together from my meeting notes, Slack, and Gmail without parsing every single artifact myself.
These tasks take time, even for the AI agent. I’m not engaged in a rhythmic back-and-forth anymore. Tasks take real time. The agent is chewing on context, and the chewing is the slow part.
This means I have a new category of time on my hands: the agent wait. And the agent wait is particularly weird, because it’s too long to stare at the progress indicator and, often, too short to pick up anything with real context-switch cost. I end up floating in the middle of my own workflow while a piece of it happens somewhere I can’t see. 🧑🚀
Early on, I watched, rapt, while the agent narrated what it was doing.
As I tasked my agent with more complex work, I started to fill the wait with Slack and scrolling. Trap. The whole reason I handed the task off to Cowork or Cursor was to spend less of myself on it. Instead, I was ending the day more drained than before, because the reclaimed time was cognitive junk food.
So, I present to you a menu. It’s sorted by wait length, because a 20-second wait and a 10-minute wait are not the same animal.
The blink (10 to 30 seconds)
Single tool call. Maybe a quick read. You cannot leave your chair. You can barely leave your current thought. But you can still do something better than stare.
- Re-read the prompt you just sent. Half the time you spot the ambiguity before the agent does and save yourself a loop.
- Unhunch. Notice your shoulders. Lower them.
- Take one real breath. Not a yoga breath, just one. 🫁
- Start drafting the next prompt in a scratch doc so it’s ready to go the second this one returns.
- Actually drink your water. 💧
The pause (1 to 3 minutes)
This is where the fast-ish runs live. A focused synthesis, a first pass, a targeted search across the context you already loaded. Enough time to do one small thing and get back before it returns.
- Open the agent’s last output and look at it with fresh eyes. The one you already skimmed and moved on from. There is a thing in there you missed. 🧐
- Capture the loose thought that just floated through your head. The one that was about to leave.
- Move one task to done in your task list. Small, stupid, satisfying.
- Refill your empty water bottle. Hydration is important!
- Send the one-sentence Slack you have been drafting in your head for two days. Heads-up, thank-you, “pushing this to Monday,” whatever.
- Close five tabs. You know which five. 5️⃣
- Read one paragraph of that article you saved. One. If it deserves more, bookmark and come back.
The breather (3 to 7 minutes)
This is a real pocket. Enough time to do one thing end to end. Do not waste it on Slack.
- Open the email you have been putting off. The one that needs more than “thanks” but less than a meeting. Write it now, while the agent is doing the actual hard thing.
- Jot down what surprised you in the agent’s last output. Not what was wrong, what was unexpected. That is where you are learning the shape of this tool. 🤯
- Plan the verification for what the agent is about to hand you. If you know how you will check the output before it arrives, you will not be tempted to squint, nod, and hope.
- Check in on a teammate. Not a status ask. A real “how are you doing” in DM. 🤲
- Add to your running list of questions you need to ask the project lead, subject matter experts, or the tech lead. You have three. Write them down.
- Go to the bathroom. 🚽
- Stand up and look at something more than 20 feet away. Your eyeballs will thank you. 👁️🙏👁️
The actual break (7 to 12 minutes)
Sometimes you hand the agent a real project. A backlog with two years of history. A synthesis across a dozen meetings. A refactor that touches half the repo. Do not waste this. It is the closest thing to a free break you get in a workday.
- Step outside. Sun on your face if you can get it. 😎
- Eat something real. Not a granola bar held over your keyboard. Food you sit down for. 🍝
- Unload the dishwasher. I know. But you will feel better.
- Do the daily mini. Connections, Strands, whatever your puzzle of choice. One. Then close the tab.
- Tidy the desktop, physical or digital, whichever is worse today.
- Walk the perimeter of wherever you are. Inside the house, around the office, lap of the block. Move your human suit! 👔
- Write down one thing you are proud of from this week. Imposter syndrome will show up eventually and the receipts will help.
The real pattern
The best thing you can do with the wait is stay one step ahead of the agent. Review the last output. Queue the next prompt. Plan the verification. Name what you actually want out of the current run before it finishes and tempts you to uncritically accept whatever it hands back.
The second best thing you can do with the wait is rest. A real rest, not an Instagram rest.
The worst thing you can do is doomscroll, because you come back from the wait more tired than you started, and now you have to review a piece of agent output with a depleted brain, which is exactly the situation where you miss the thing that matters.
Your agent is running. 🏃 You have somewhere between 10 seconds and 12 minutes. What are you going to do with it?