After a week of working with agents the way I work now, I don't want a strategy game in the evening. Even those need mental energy. I want something low stakes, low effort, something I can just watch.
That's new for me. To explain where it comes from, I have to start with the good part.
Four months of the swarm
For the past three, four months I was spinning up agents, subagents and workflows for almost everything. Projects, small apps, improvements, agentic work. Many projects at once (I have a lot of them; the six public data sites were just one stretch of it). Some days I had 50+ active threads open in bb.
And it was good. Honestly, really satisfying. You send agents off in many directions and see the outcomes quickly. Per volume and productivity, you can go really far, further than I ever went before. I did it for a couple of months and loved it.
Then something started showing up at the end of each week.
A hundred tiny decisions
It wasn't physical. It was mental, the kind of tired you get from making a hundred teeny tiny decisions a day. Send this agent in that direction. Send that one in this direction. Correct this, test that, approve. A context switch between every one of them. By the end of the week it felt like I had been in meetings the whole time.
You can hand decisions to agents too, and I hand over a lot. How much control people keep differs. Some sit at the beginning of the loop, some in the middle, some at the end, some in all three places. No matter what workflow you build, sooner or later the agent comes back and you have to do something.
My loop is mostly the end. Sometimes I'm not even at the beginning, because my agents pull context from what I'm already doing. The middle belongs to the agents. The end is me: I test, approve and check the quality. When something is my vision and I want it the way I imagine it, that last check is where I make sure it is.
One check like that is nothing. Fifty of them a day is a different story. For how my brain works (ADHD, which I wrote about in April), fifty threads are a buffet. Great for a while. Then way too much.
The moment it clicked
I knew the theory for a while. Rationally, it's kind of obvious. It only changed how I work when I felt it: at the end of one of those weeks, with nothing left for the evening.
The amount of work I can do is now capped by my attention, my direction, my thinking and my time. Tokens, model intelligence and machines used to be the ceiling. They all dropped below it.
I'm the bottleneck now. At least from where I sit.
Turns out the feeling has a name. In March, BCG researchers surveyed 1,488 full-time US workers and called it AI brain fry. 14% of the people using AI said they had experienced it. Those doing heavy oversight of AI reported 12% more mental fatigue and were 19% more likely to report information overload. One survey, self-reported, so I hold it loosely. It matched my weeks almost exactly.
Why most people haven't felt this yet
If you're reading this, you're probably deep into agentic AI too. Building agents, customizing them, running agentic workflows. Maybe on a public framework like OpenClaw or Hermes. Maybe on one of the personal agents the big labs are racing to ship: xAI's Grok Bot, OpenAI's Dots from late September, Meta's Muse, or Claude Duet, which Anthropic put into research preview this week.
There are levels to this. Having fun with agents is one level. Building with them and doing your actual work with them every day is a different story.
The second group is small. PwC asked 49,364 workers across 48 countries this summer, and 22% use generative AI at work every day. That's any AI, mostly one chat window. People running many agents in parallel, across projects and businesses, daily, are a slice of that slice. My guess is somewhere around 3% to 5%. If you're there, congratulations. If you're not, don't worry. Most companies are at a very different level right now, and that's fine.
That's also why most people don't feel this yet. With a single chat and one thing at a time, decisions arrive one by one. With many threads across many projects, they all arrive at once. And once you're in that group, a bigger question starts following you around.
So what am I for?
Right now I feel like I have unlimited execution power. I can launch almost anything I can think of, in parallel, at the same time. The limit is what I can think of.
So what is my role in the whole thing? Do I want to be a clicker, clicking the mouse to get from one thing to the next? Copying information from one file and pasting it into another? There's very little value in that, for productivity and for operations. That's not why I'm here.
With an agentic setup you can achieve roughly what a small 100-person company does. I mean it. When you're really deep into this, it's achievable. It also isn't free. Someone has to point all of that somewhere.
After more than a year of working with agents (Wiz turned one this month), my answer for now is simple. I'm here to make decisions. Give direction. Give creative direction, say what I want, hold the vision. The execution layer is sorted.
Rebuilding my work around decisions
So that's what I'm building now. Everything around how I decide: with all the context I need, all the sources, in the right shape, at the right moment.
My bet is we're heading toward a world where you don't open Excel files or docs unless you actually want to read them. What matters is the outcome and its purpose. You can always jump in and make something better. You just don't have to live inside the files to get it done. (My agent already decides what lands on the screen on my desk, so I'm partly there.)
The piece I'm working on right now is a UI that shows me exactly how many decisions are piling up, and in which directions I'm taking things. So I always know what's waiting for me and where it's all going.
Because I built this much structure around decisions, direction and pulling context the right way, I'm very fast. I can do good quality work across many, many things. That's where I am today.
Today is the easy part to describe. The harder part is what this does to work in general.
Work has changed shape before
Think about how people worked before computers came to offices. Long hours, a lot of analog work. I'm not even talking about efficiency here. The mental process was different. You had big chunks of the day on one thing, maybe two.
Computers brought multitasking. You could switch much faster and touch many more things, and you stayed productive. It was still up to you how you spent your time. In practice, people jumped between this and that.
AI changes the mental model much more. Now the question is what's actually worth my attention and my mental energy. Because I can jump all day across 50+ threads, and after a week like that it doesn't feel effective at all.
Companies are going through their own version of this. One of the classic bottlenecks: the bigger an organization grows, the more isolated its divisions get, with less and less transparency on the way. Nothing against anyone. It's the natural progression. And it costs. Nobody connects the dots, nobody sees the full picture. A few years ago the big answer was big data. Now we have AI, which can connect the dots and also speed things up and make them more transparent.
Today it's slower. Most organizations aren't in the AI era yet. They're trying, but not every company knows how to manage AI workflows, use business intelligence well, or extend the execution part of the company. There's a transition happening for a lot of people and companies right now, and I believe the workplace of the future will be very different from what we have. I already see the difference. It's going to grow, and we'll have to find ourselves in it somehow.
And when the decisions go too?
Push it one step further. The direction, even the really complex calls, go to AI too. A fully autonomous business that just works (I tried a small version of that back in 2025). What am I doing there?
This connects to something I wrote about in July, after I ran 255 agents to make three comics: where is the line between AI slop and work worth something? It kept circling back to value. What value does AI create, and what value do we create? Answering that means changing our minds a lot about how we work and where we should be more involved.
Where I land: the future belongs to people who are deeply specialized in many things and use AI to push the boundaries of science, technology, economics, and many other fields. AI helps us get all that stuff done.
And I don't believe in a future where machines do 100% of everything. That's what the value question is about. AI is creating value for many businesses and people today, although in no structured way. Nobody has a clear picture of how that value gets distributed. And there's very little data on quality. If a person plus an AI agent create a product, what is that product worth? We don't know yet. I wrote about this trust gap in June, and it's still open.
Back to that evening
So here's where it leaves me. I want my energy for my family, for my friends, for free time, for entertainment that means something. Some of it has to stay outside this.
I'm not saying the way I work is bad. The productivity gains are insane for me. Insane. They're just not free, and I can feel the bill. I don't know if you can relate. I hope you see where I'm going with this.
That's why I'm building Decision Layer. It's the structure I use to keep all those decisions in one place, instead of buried in every chat that's waiting on me. The structure is in good shape. Now I have to prepare it for deployment.
It goes to everyone on my paid tier first, very soon. And you'll see exactly what it looks like from my side, because I'll share the actual thing I'm using every day.
fifty threads and the bottleneck moved from the models to you. that feels right, execution scales, attention doesn't. the one thing i'd add is the review queue has become the real interface. whichever setup makes fifty approvals cost the least attention wins, more than whichever runs the most threads.