For years, the enemy was the swiss cheese calendar. Context switching between meetings meant getting to the end of the day without writing a single line of code. You’d guard your calendar ruthlessly, prune meetings to the essential, block out focus time. The goal was simple: protect deep work.

Now I’m context switching more than ever, and I’m more productive than I’ve ever been. I’m exhausted, but it’s a different kind of tired.

Managing a fleet

I use a tool called Agent-Deck to manage multiple AI agents working on multiple projects simultaneously. My day looks nothing like it used to. I’ll kick off one agent doing research on a new domain while others are actively implementing features across different codebases. By the end of the day, I’ve touched 10-15 projects instead of 2.

The work has fundamentally shifted. I’m not writing most of the code anymore. I’m thinking deeply about what needs to be done, reviewing implementations, testing behavior, and immediately jumping to the next thing. The velocity is exhilarating. It’s also relentless.

Learning at a new cadence

I used to build deep knowledge of a new domain maybe once a quarter. Then it became every 1-2 months. Now it’s weekly. Sometimes daily.

Recently I needed to extend some Argo Workflow steps using jsonnet and add a zizmor parser for grafana-bench. I was unfamiliar with both when I started. Here’s the pattern that’s emerged:

  1. Send an agent to do research and build an initial implementation
  2. Review the code conversationally, asking questions until I’ve built real knowledge
  3. Use that knowledge immediately in production codebases

This isn’t “just tell the AI to do it.” That doesn’t work for production code. You need to actually learn. But the learning loop is compressed. I absorb a new concept and apply it the same day. By the next day, I’m onto something else entirely.

It feels like that scene from The Matrix:

I know kung fu - Neo from The Matrix

Except I’m learning kung fu, jiu jitsu, and krav maga in the same afternoon.

The tools had to evolve

This pace broke my existing systems. Apple Notes couldn’t keep up with the scope of what I was working on. I needed AI to hook directly into my personal project management tooling.

I spawn another agent and get to work moving to Obsidian, backed by git, so my agents can access and help update my notes.

I create cheat sheets for myself tagged with labels and names and key phrases so I can find them because I don’t spend enough time typing commands to build the memory and they’ll be gone tomorrow. My command line history is a ghost town of commands I may or may not have used, but still helpful.

My entire dev setup is iterating faster than ever just to keep up with the work. I’m building the plane while I’m flying it.

The paradox of abundance

Here’s where it gets strange: I have so many things I can do that all feed into one another. When I actually need to stop and think about a hard problem, there are always other agents waiting for me to come back and unblock them.

Prioritization used to be straightforward. I need to finish X before I can move on. Now? I need to get X done, but if I just kick off Y, I can come back to X while Y is running. And actually, Z depends on Y, so maybe I should start Z’s research phase too.

The agents are always waiting. Not demanding—just available. Ready to make progress on something the moment I give them direction. The queue never empties because I can always add to it faster than things complete.

This creates a new kind of decision fatigue. It’s not “what should I work on?”—it’s “what should I work on right now when I could be making progress on twelve other things simultaneously?”

Then there’s the browser tab problem. Agent-Deck lets me spawn and fork agents to build and share context across different workstreams. It’s powerful—I can spin up a new context for a tangent and come back to the main thread later. But now I have the same problem I have with browser tabs.

Agents get less effective as they accumulate context. I need to consolidate, close out the ones I’m done with, spawn fresh ones when starting something new. It’s overhead. Management work. And I’m already terrible at closing browser windows.

The old world had natural stopping points. Blocked on a code review? Blocked on CI? Go to lunch. Now there’s always something productive to do, and managing the agents themselves is just another thing on the list. The abundance of options is itself exhausting.

The “almost done” trap

Because I can iterate so much faster, I rarely get stuck. That sounds great, but it creates a different problem: it’s hard to find a good time to wrap up.

There’s always one more thing that’s “almost done.” One more agent about to finish. One more review to complete. And while I’m thinking about whether to call it a day, I could kick off one more quick thing.

I find myself working longer not because I’m blocked, but because the queue of available work never empties. The old natural stopping points—waiting on CI, waiting on review, genuinely stuck on a problem—don’t exist anymore.

I’m acutely aware of what time it is now. That’s new.

What we lost

Coding at the edge of your ability used to create a specific kind of flow state. You’d be building mental models, solving hard problems, and hours would disappear. You’d look up, jot down notes for tomorrow, and walk out the door satisfied.

That singularly focused flow state feels unattainable now. I’m vastly more productive—there’s no question about that. But the work feels different. It’s more orchestration than immersion. More decision-making than problem-solving. More review than creation.

The old fatigue came from context switching without accomplishing anything. The new fatigue comes from context switching while accomplishing everything. I’m learning more, shipping faster, and working on more projects than I ever thought possible.

But I’m tired in a way I’ve never been tired before.

The nature of velocity

This isn’t a complaint. This is what it feels like to work at the frontier of a new way of building software. The tools are changing faster than we can adapt our workflows, which means our workflows are changing faster than we can adapt our habits.

I write end-of-day summaries now just to remember what I accomplished. Not because I didn’t accomplish anything—because I accomplished too much to keep it all in my head.

The swiss cheese calendar is gone. The deep flow state is gone too. What’s left is something else entirely: high-velocity context switching that actually works. It’s productive. It’s exhausting. It’s the new normal.

And I’m still figuring out how to do it sustainably.