The New Deep Work — Macro<br>July 16, 2026<br>The New Deep Work<br>I was doing some customer support in Sunsama the other day, and Shaheed had written in asking me how to deal with being sucked into his coding agents, forgetting other responsibilities, and feeling bad about multi-tasking. He put it better than I’m summarizing:

I used to use Sunsama every day and it became part of my ritual. However, recently I have been working more and more on coding projects. For that, I am sitting in Claude Code all day and going back and forth responding and making decisions. As a result of having to wait for agent responses, I have started working on multiple projects at the same time. I know that this means that I have to context switch a lot and that’s not great for productivity… I used to be really structured and planned about my days but now I just find that I work on coding related projects and basically neglect other responsibilities. I feel like I have lost discipline… I’m working on 3 repos at the same time on some days. I thought about doing 1 repo per day but then I ‘feel’ like I am not making progress. Do you have any tips on how I can get back on the bandwagon?

It got me thinking about the “new deep work” — how I’m actually working now, how I stay sane with a bunch of agents running, and if multi-tasking is really that bad.

I think we’re all figuring this out together right now. We’ve all felt like Shaheed in the past few months. Here’s what I told him, which ended up being a mini-essay on the new deep work that I thought I’d share broadly.

I want to first address the meta point on context switching being bad for productivity.

I’ve long been a proponent of deep work as the way to do high-quality work in a way that feels good. It’s wonderful to have your attention fully immersed in something, and it’s best if it’s one thing. I am not sure if that still holds up. If you were to truly single-track, you’d sit around twiddling your thumbs or browsing the internet while your coding agent worked. It’s definitely not fulfilling. I think the new deep work is to immerse yourself in building. Does it mean running 11 parallel agents and feeling scatterbrained and confused? Probably not. Does it mean there are times where you’ve got 2-5 agent threads open that you are driving? Probably, yes.

I don’t know what the limit is, but I do know it’s contextually dependent. If I am feeling fresh, motivated, super into what I am working on, and the conversational depth isn’t too hard, I can legitimately work on 5 agents at a time. But things vary. On days I’m tired, or I work in the evening, I definitely will just run one agent and watch some TV in the background. And then of course the nature of the work matters too. The other day, I kicked off three big refactors. I was just giving my agent big-picture guidance and not thinking hard. When I get into the final details of testing and feeling things out, I need to spend one-on-one time with an agent. Not all jobs are the same. A big feature, versus a bug fix, versus a feature that requires some design work, versus one that just slots right into an existing architecture. My parallel agentic usage ebbs and flows with the kind of work I am doing.

I am not single-track focused anymore. I felt bad about it at first, like I was violating some great principle of productivity, but it helped me realize that what I care about is being immersed in my work.

Now, let’s get into some more practical strategies for staying sane in this new world (I’ve also found myself lost to building things and not wanting to answer emails anymore).

Here are some techniques I use.

I use my daily task list as a list of “deliverables.” It’s too easy to go on side quests. I need my task list to remind me what I want shipped live today, who needs a response, what paperwork must be done. AI coding definitely makes you feel like you are productive, but real productivity is doing the things you think truly matter. If you don’t define those to start the day, you are just going to be on a meandering journey of blasting out code all day.

I do tasks in groups. Instead of setting aside time for bug A and bug B, I set aside an hour to rip on any bugs in a certain category. Then I will do 3 hours of building a big feature. While I build a feature, I will have one agent working on the main job and others working on small adjustments and polish related to it. This gets me that “deep work” feeling. I am ostensibly focused on the “Sunsama Daily Autoplanner,” but that is actually me building one core flow and fixing a bunch of small papercuts I noticed along the way.

I work on my next task when a current task is finishing. In fact, I am writing this while I’m ostensibly waiting for another task to finish (but the remaining work is out of my hands). I never used to do this before. I do this all the time now. The task I currently have my timer running on is my current “top priority.” When my agent gets back, I stop and give that...