The Ceiling

I have never, in my entire life, hit a cognitive ceiling. Not until a few days ago.

I've been the person with thirty things going at once for as long as I can remember. At work, at home, in my head. Multiple projects, multiple priorities, multiple conversations, all running at the same time. That's not a complaint. That's how I'm wired. I thrive in the chaos. I don't just tolerate it. I run toward it. I've built a 20-year career on the fact that I can hold more context in my brain than most people think is reasonable, and I can switch between it all without dropping anything.

Or so I believed.

Let me paint the picture.

On any given workday, I have somewhere between 12 and 18 Claude Code sessions running simultaneously. Each one is a different project. A different system. A different set of decisions. At home, I've got another 10 or so open. Personal projects, things for friends, things for my family, things I'm building just because I want to see if I can. Each session moves at its own pace. One stops and asks, "Do you want to proceed?" Another needs me to authorize something. A third is waiting for a decision about architecture.

I work through them the way you'd read a book. Top left to bottom right. I scan each terminal, read what it's doing, make a call, and move on to the next one. Line by line by line. Round robin. I invented that system because there was no better answer for how to prioritize when everything is asking for your attention at the same time.

My output is probably triple what a normal person produces. The day I'm writing this, Claude and I created, worked, and closed 30 ADO tickets at work. That's not counting the ones already open that I kept working through from the plan. In my personal life, I'm moving through things that would normally take an entire team, and I'm doing them part-time as I get the context to get to them.

And at the end of the day, what I feel is exhaustion.

Not tired. Not sleepy. Not the kind of done where you put your feet up and feel satisfied. Actual, genuine brain exhaustion. For the first time in my life, my brain hurts.

A few days ago I watched a video by Matt Wolf called "AI Is Frying Your Brain." He talked about feeling the same thing. More productive than ever, more drained than ever. He talked about struggling to come up with ideas for videos after outsourcing that process to AI for too long. And he pointed to research that I spent the next two days reading.

It turns out there's a name for what I'm feeling. Harvard Business Review just published it. They call it "AI brain fry."

Boston Consulting Group and UC Riverside surveyed nearly 1,500 full-time workers and found that the most mentally taxing form of AI use isn't generating content or brainstorming ideas. It's oversight. The act of directly monitoring what AI is doing. Workers who reported high levels of AI oversight spent 14% more mental effort, experienced 12% more mental fatigue, and reported 19% more information overload than those who didn't.

That's me. That's exactly me. I am the person monitoring 12 to 18 active AI sessions at once, reading their outputs, making micro-decisions, correcting course, approving direction. I am the definition of AI oversight at scale.

The study found something else that stopped me cold. There's a clear productivity inflection point around three simultaneous AI tools. One tool, productivity goes up. Two tools, it goes up more. Three tools, it peaks. After three? It drops. Because the human brain can't coordinate more than that without the cost of switching eating the gains.

I'm running six times that number.

There's an engineer named Siddhant Khare who wrote an essay called "AI Fatigue Is Real and Nobody Talks About It." He described the exact paradox I've been living. AI makes each individual task faster. What used to take three hours takes 45 minutes. But instead of doing fewer tasks, you do more. Your capacity appears to expand, so the work expands to fill it. Your manager sees you shipping faster, so the expectations adjust. You see yourself shipping faster, so your own expectations adjust. The baseline moves.

Before AI, he said, he'd spend a full day on one design problem. Sketch on paper. Think in the shower. Go for a walk. Come back with clarity. One problem. One day. Deep focus.

Now he touches six different problems in a day. Each one only takes an hour with AI. But context switching between six problems is brutally expensive for the human brain. The AI doesn't get tired between problems. We do.

He nailed the paradox in one sentence that I haven't been able to stop thinking about: AI reduces the cost of production but increases the cost of coordination, review, and decision-making. And those costs fall entirely on the human.

That's my life. Every single day. The production is effortless. The coordination is where the weight is.

But here's where my version of this story gets different from Matt's and Siddhant's and the HBR study.

They all describe this as something that happened to them. Something they noticed creeping in. Something that started subtle and got worse.

For me, it's not subtle. For me, it's a full-speed collision between the thing I've always been and the thing AI just showed me I can't be.

I wrote a post a few weeks ago about being a jack of all trades. About how the thing I was once insulted for turned out to be the exact wiring the future requires. I meant every word of it. I still do. My ability to hold context across domains, to move between problems, to understand enough about a dozen things to steer a model through any of them, that is genuinely valuable. It's what makes me good at this.

But what I didn't write in that post, because I hadn't felt it yet, is that there's a ceiling. The jack of all trades can hold more plates than anyone in the room. But AI just handed me 30 plates, and it's handing me more every day. And somewhere between plate 18 and plate 25, the advantage stops being an advantage and starts being a liability. Because a person who's wired to say yes to everything, who's driven by value above all else, who measures her worth by what she delivers, that person doesn't know when to stop.

I've never had to know when to stop. There was never enough capacity to make stopping a question. I could always handle it. I could always do one more thing.

Now I can't. And I don't have a framework for that.

In the last couple of days, I've started noticing things. I double-read prompts. I wonder, "Wait, what was that session doing?" I put the wrong response into the wrong Claude Code terminal and it looks at me like I've lost my mind. My downloads folder at work is a disaster. Files on top of files on top of files. It's so easy now to throw a hundred thousand lines of code at something that you don't think about the technical debt you're building. Projects happen so fast that my brain can't form a proper memory of them.

So I built a system. Personalized markdown files for every project. Who, what, when, where, why, and how. So that when someone comes back and asks for a change, I can catch my own brain up on something I built three days ago and can't fully recall. I'm teaching myself with markdown. I am literally writing documentation for my own memory because my brain can't keep pace with my tools.

That's a sentence I never expected to write.

And then there's the guilt.

I wasn't expecting this one.

When you have 10 sessions waiting for an answer, you feel guilty for not being fast enough. I look at a terminal that's been waiting for me and think, I'm delaying this. I am causing AI to be slow because I can't get to it quick enough. That's a completely new kind of guilt. The guilt of being the bottleneck in your own workflow. The guilt of holding back the thing that's supposed to make you faster.

And with that guilt comes the prioritization problem. When everything is asking for attention, how do you choose? I chose round robin because there was no better answer. But I'm not actually round-robining. I'm skipping the passion projects because they can wait. I'm triaging by urgency, not by joy. And the things that energize me, the things I build for myself, they're the first to get pushed aside.

I never had to prioritize like this before. Not at this scale. Not at this speed. I don't have a framework for choosing between 28 active projects, and the absence of one is itself a source of exhaustion.

I want to be clear about something. I don't feel like I'm getting dumber. I don't feel like my brain is shrinking or that I've lost the ability to think. I'm as sharp as I've ever been. What I feel is that I've found the edge of how much I can hold at once. The line between what I think and what my tools think has blurred in ways I've written about before, and I'm so deep in the work that context bleeds across sessions, across projects, across the boundary between my job and my life. It's not that the thinking got worse. It's that there's more thinking than any one person was ever meant to do.

That's the ceiling. Not a loss of ability. An excess of demand.

So what's the answer?

I don't fully have one yet. But I know the question is wrong when people ask, "How many things can you do at once?" The right question is, "How many things should you?"

The BCG study says three simultaneous AI tools is the inflection point. After that, productivity drops. I'm running 12 to 18. Somewhere between three and my number is a sustainable zone that I haven't found yet.

Siddhant's advice is good: time-box your sessions, separate thinking time from AI time, accept 70% from the model and stop chasing perfect. Matt Wolf started writing on paper again, away from his computer, to exercise the thinking muscle. The HBR researchers say organizations need to treat human attention as a finite resource and design AI workflows with the biology of the brain in mind.

All of that resonates. But I'm not the kind of person who reads advice and nods. I'm the kind of person who builds a system.

So here's what I'm doing. Starting next week.

I already mentioned the markdown files. Every project at work gets a personalized markdown with the who, what, when, where, why, and how of that project. Because the work happens so fast now that my brain doesn't have time to form a lasting memory of what I built, those files are how I catch myself up when someone comes back with a change request three days later. I'm also adding incremental session updates and memory notes to each session as it progresses, so the context stays current and I'm not reconstructing from scratch.

But the bigger change is the cap. I'm limiting myself to eight concurrent sessions next week. Eight at work, eight at home. Not 12. Not 18. Eight. I want to see if that number lets me stay in the zone without the exhaustion.

I'm going to arrange my windows around priority. The four most important sessions go center of my main screen, right where my eyes land first. The remaining sessions arrange around them, with the most important of what's left on the left side and the least important on the right. If I have a major project running at work, I'm going to minimize the low-task sessions, the research, the innovation experiments, and keep the priority work front and center. And I'm committing to something I've never done before: if I need something from another session, I will finish what I'm doing in the current one first. No more bouncing. No more round robin. Sequential attention within the cap.

Every morning, my pre-planning is going to include a framework pass where I decide which sessions are the most important for that day. Not which ones are loudest. Not which ones have been waiting longest. Which ones matter most.

At home, I'm taking a different approach for the sessions that don't actually need me. My security agent is a good example. It runs through all of my code, does security reviews, outputs a report to a folder that my other agents pick up and act on in their own sessions. It's scanning code, looking for vulnerabilities, checking the internet, making suggestions. I don't need to approve every step of that. So I'm going to let it run with dangerously skipped permissions and trust the pipeline. Same with my personal research sessions, the ones working on blog post research, articles I need to read, gift ideas for my family. Those don't need me in the loop giving permissions for every search. I'm going to let them run autonomously and review their output when they're done.

The distinction I'm drawing is between sessions where I'm the subject matter expert making real decisions and sessions where I'm just a bottleneck granting permissions. If a session doesn't need my judgment, it doesn't need my attention. Freeing those up means the eight sessions I am running get the brain they deserve.

And the last thing I'm building is a "session learnings" system in my PARA setup, a shared knowledge base where critical learnings from any session are stored so that all of my sessions have access to what the others have discovered. Cross-pollination without me being the courier.

Here's the realization that pulled all of this together.

I have spent months building context and token management systems for my AI sessions. I've thought deeply about how to keep my models on track, how to manage their memory, how to make sure they have the right context at the right time so they can do their best work. I've built entire architectures around the idea that a model performs better when its context is clean and focused.

I never once did that for my own brain.

I've been managing context for my tools while letting my own context run wild. Eighteen sessions open. No prioritization framework. No memory system for myself. No cap. No boundaries. I gave my models everything they needed to think clearly and gave myself nothing.

For people like me, the ceiling isn't just a performance problem. It's an identity problem. Because if I can't do all the things, who am I? I've spent my whole life being the person who could. AI made that feel limitless for a while. And now it's asking me a harder question: not what I'm capable of, but what's sustainable.

I don't have the full answer to that yet. But I think treating my own brain with the same care I give my models is where it starts.

I'm not writing this to scare anyone. I'm not anti-AI. I have never been more in love with the tools I use or more amazed at what they let me build. I closed 30 tickets the day I started writing this post. I'm building systems that would normally take teams. I'm living in the future and I wouldn't trade it.

But I'm also exhausted in a way I've never been exhausted. And if someone who has done this her entire life, who was built for this kind of work, is hitting a wall, I think it's worth talking about. Because a lot of you are about to walk the same path I'm on. And you deserve to know that the ceiling is real, that the science says it's real, and that hitting it isn't a weakness.

It's just what happens when the most powerful tools ever built meet the only brain you've got.

Take care of yours.