My ADHD brain runs 20 AI agents a day. Here's what happens.

I’m dyslexic. I show a lot of ADHD traits, though I’ve never had a formal diagnosis. Ideas arrive fast. Attention shifts faster. I work in bursts, not blocks. I’ve been like this my whole life, and for most of it, the tools I had to work with punished that.

Then I discovered agentic AI. Not chatbots. Not “ask a question, get an answer.” I mean proper AI agents that hold context, run tasks, and work alongside you. It started with OpenAI’s Codex, then Claude Code, then tools like Cowork and GitHub Copilot agent mode. I now run 20 terminal windows simultaneously, each one holding context for a different project, a different thread, a different half-finished thought my brain dropped 40 minutes ago.

For the first time in my career, the way my brain works isn’t fighting the system. The system finally works like my brain.

But I’ll be honest: I’m not sure this is entirely healthy. And I think that’s a conversation worth having.

How my brain actually works

I’ve never been formally diagnosed with ADHD. These things are all spectrums, and honestly I don’t feel a special need to pin down exactly where I sit on this one. Dyslexia was different. That had me properly struggling at school and then at uni. I got tested for it in my first year at Newcastle and it genuinely made a difference, suddenly things made sense. But ADHD? I’m 46 this year. I know a thing or two about myself by now, how my brain works, how I cope, and what happens when I don’t.

I’ve built a marketing agency over 11 years, worked with Microsoft, Google, SAP, TCS, and hundreds of B2B tech companies. By any external measure, things have gone well. But internally? It’s not always been chaos. Sometimes things run smooth. But it’s always been full on. Ups and downs. Luckily I’ve always had good people around me, at home and at work. But yeah, it can be a lot.

My brain can do “sit down and work through this methodically for three hours.” Sometimes. But it takes real focus to get there, and it’s not the default setting. I have a mix of working styles, but I tend towards more things rather than fewer. It’s more like “have seven ideas before breakfast, hyperfocus on one for 90 minutes, completely forget the other six existed, then remember them all at 2am.” I misspell things constantly. I lose track of conversations mid-sentence. I can build an entire analytics dashboard in a morning but forget to reply to the email that started it.

For years, I compensated. Lists. Reminders. People around me who held the threads I dropped. Asana. God, Asana. I used it religiously for years, and so did the whole company. Every personal task, every project, everything tracked. I knew we’d taken it too far when a project manager created a task called “Be Creative”, assigned it to three people, gave it a date and a 90-minute time slot. Madness. But it worked, mostly. The work itself was never the hard part. It was the overhead of managing a brain that doesn’t come with a filing system.

What running an agency with this brain looks like

Before I get to the AI bit, I want to be honest about what it’s actually like running a company when your brain works this way. Because it’s not all upside.

The good parts are genuinely good. New ideas come constantly. I’ll see a pattern across three client conversations and have a product concept sketched out by lunchtime. The agency has always been early on things, whether that was AI tools, GEO, or building our own SaaS products. That restless “what if we tried this?” energy is what built Compass, what got us to 3,200+ client opportunities, what keeps us pushing forwards while other agencies play it safe. Innovation isn’t something I schedule. It’s just how the brain works.

The bad parts are harder to talk about. I know what my team sometimes thinks, even if they’re too kind to say it. Another new idea from Chris. Another change in direction. Another Monday morning where everything we agreed on Friday gets rearranged because he had a thought over the weekend. From the outside, it can look like a lack of strategy. Like I don’t know what I want. The truth is the opposite: I can see too many things at once, and I want to chase all of them.

I’ve learned to manage this better over the years. I have good people around me who push back, who ask “is this actually a priority or is this Tuesday’s idea?” That honesty is invaluable. But the tension between “visionary founder” (yes, I said visionary) and “guy who changes his mind a lot” is something I navigate every single week.

Does AI change how we work?

Before I get into the 20 terminal windows thing, it’s worth talking about what AI tools actually do to the rhythm of work. Because it’s changed something fundamental.

Large language models generate tokens. That’s all they do. You type a prompt, the model produces a response, one token at a time. Sometimes it’s fast. Sometimes you’re sat there watching it think. Then reasoning models came along, models that “think” before they answer, and the responses got better. But the waiting got longer.

Take it further with agents. Tools like Claude Code, which I use constantly, don’t just think. They reason, they use tools, they read files, they write code, they test things. A single task might take minutes of autonomous work. You fire it off and then… you wait.

So what are people doing while they wait for AI to finish?

It’s an interesting new paradigm. Some people sit and watch. They want oversight, they want to check every step. Fair enough. I imagine others read the BBC while their agent chugs away. But for me, and I know I’m not alone, the waiting time is when I fire something else up. Then something else. Then before I know it, I have multiple agents running different tasks across different projects. Suddenly I’m commanding an intelligent AI army doing genuinely useful work across the business.

Or I’m spinning plates maniacally. Depends on the day.

The point is: AI hasn’t just changed what we can do. It’s changed the tempo. The gaps between actions used to be dead time. Now they’re launch windows. And if your brain already works in parallel, if you’ve spent your whole life bouncing between things, this new rhythm fits like it was built for you.

What 20 terminal windows looks like

Right now, as I write this, I have Claude Code running in 20 separate terminal windows. One is building out an outreach system for PR targets. Another is fixing authentication bugs in our analytics dashboard. A third is monitoring our workstreams, checking which emails got replies, updating trackers. There’s one researching a client’s competitors. One refactoring a script I half-wrote last week.

Each agent has its own context. It knows what the project is, what’s been done, where I left off. When I switch windows, the agent doesn’t need me to explain what I was doing. It already knows. It picks up exactly where my attention wandered off.

This is the bit that’s hard to explain to people whose brains work differently: the problem was never intelligence or capability. It was continuity. The ability to hold a thread across time. To pick something up on Wednesday and remember what you were thinking on Monday. Research suggests this isn’t purely an ADHD thing. Working memory limitations affect everyone. But if your brain already struggles with executive function, the impact is amplified. You don’t just lose the thread, you lose the thread and three other things you were meant to be tracking at the same time.

AI agents are external working memory. That’s not a metaphor. That’s literally what they do for me. I often come back to a tool after hours or days away, and the first thing I type is “remind me what we were doing.” And it does. Perfectly. Every time. Not only can AI do more for us, it never forgets and it can bring us back up to speed in seconds. For someone whose brain drops context the moment something shiny appears, that’s not a nice-to-have. It’s transformative.

I’ll spin up an agent, dump a half-formed idea into it, switch to something urgent, come back two hours later, and the agent has the full context sitting there waiting. It didn’t get distracted. It didn’t forget. It didn’t move on to something more interesting. It just held the thought.

The tooling has evolved fast. Codex showed me that AI could write code alongside me. Claude Code showed me it could hold an entire project in context and work autonomously. GitHub Copilot agent mode, Cowork, and other agentic tools mean I can have multiple agents running simultaneously across different codebases, each one as a persistent collaborator, not just a one-shot helper. The combination is what makes 20 terminal windows possible. No single tool did this. The ecosystem did.

The superpowers nobody talks about

There’s a stat that neurodivergent employees are 55% more likely to use AI than their neurotypical colleagues . That doesn’t surprise me at all. We gravitate towards these tools because they solve problems we’ve been fighting our entire lives.

But there’s something else happening that I didn’t expect: AI agents actually reward the traits that traditional workplaces penalise.

The context-switching thing. Neurotypical productivity advice says “single-task, do deep work, block your calendar.” My brain doesn’t do that. It hops between things. With 20 terminal windows, each hop is productive. I’m not losing context when I switch, because the agent holds it. The cross-pollination between projects that felt like a bug in my operating system is now a feature. I notice patterns across clients and projects specifically because I’m cycling between them.

The burst-work thing. I can do 12 hours of work in 4 hours, then nothing useful for the rest of the day. AI agents match this rhythm. When I’m in a burst, I can orchestrate 20 agents simultaneously, launching tasks, reviewing outputs, spinning up new ones. The agents don’t care that I’ll be useless by 3pm. They keep running.

The spelling doesn’t matter thing. I’m dyslexic. I misspell things constantly in my prompts. Agents don’t care. They understand intent. For the first time, the gap between what I’m thinking and what I can express doesn’t slow me down. An AI that can parse “i want to buiild an anaytics dasboard that shwos refferal traffic” and just build it? That’s not a productivity tool. That’s an accessibility tool.

The dark side of infinite capability

When every idea can be instantly acted on, where’s the filter? Before AI agents, the friction of execution was a natural governor. Some ideas died because I couldn’t be bothered to build them, and that was fine. Most ideas should die.

Now I can spin up an agent and have a working prototype in 20 minutes. The dopamine hit of “new thing working” is immediate and powerful. So I do it again. And again. By lunchtime I’ve started six new things and my brain feels like it’s been in a tumble dryer.

The hyperfocus that my brain gives me, the ability to lock onto something and block out everything else, is incredible when pointed at the right thing. But AI agents make it possible to hyperfocus on orchestrating agents, which is a strange, recursive loop. I’m hyperfocusing on delegation. I’m micromanaging a fleet of AI workers with the same obsessive energy that once went into the work itself.

The World Health Organisation estimates that untreated ADHD costs about 22 lost productivity days per year. I’d argue that AI-amplified, unsupervised ADHD traits might cost different things: quality, depth, the ability to know when to stop.

My brain at 9am with 20 terminal windows open is electric. My brain at 3pm is soup.

What fried brain actually costs at home

This is the part I find hardest to write about.

I close my laptop at 5pm and the work is done, but my brain isn’t. Georgia’s telling me about her day and I’m nodding, but part of me is still replaying whether that API fix will pass the tests. Thomas wants to play Fortnite and I’m physically present but mentally still cycling through agent outputs.

Katie can tell. She always can. “You’re not here,” she’ll say. And she’s right. My body came home but my brain is still in 20 terminal windows.

It’s not all doom. I manage it. Katie’s known me long enough to spot the pattern, and I’ve got better at catching it myself. But it’s something I have to stay on top of. The line between “switched on” and “checked out” is thinner than I’d like, and AI tools, by making work more immersive, can push you closer to that line without you noticing.

I don’t have a neat answer for this. I run, which helps. I’ve got better at recognising the signs: the glazed look, the half-second delay before I respond to a question, the way I’ll pick up my phone and open a terminal without thinking about it. But I’d be lying if I said I’ve solved it. The tools that make my working brain extraordinary make my evening brain depleted. That’s a trade-off I’m still figuring out.

AI as the first real cognitive accessibility tool

Around 15 to 20% of the global population is neurodivergent . That’s not a niche. That’s roughly one in five people. And until very recently, every productivity tool, every project management system, every workplace process was designed for brains that work in straight lines.

What AI agents do, specifically the kind that maintain persistent context, respond to natural language, and don’t penalise non-linear thinking, is something genuinely new. They’re not just making people faster. They’re making a certain kind of brain functional in ways it wasn’t before.

Researchers are already framing this as a rights issue. A recent academic paper proposes recognising “Assistive Artificial Intelligence” as a fundamental cognitive right for neurodivergent people, arguing that AI systems operate as “Cognitive Navigation Interfaces” that reduce executive overload, filter informational noise, and make complexity more accessible.

That language feels right to me. When I use Claude Code, it’s not doing my thinking for me. It’s holding the scaffolding so my thinking can happen. It’s the difference between having ideas and being able to act on them. I’ve always had the ideas. Now I have the infrastructure.

So is this healthy? I genuinely don’t know

I want to end with honesty rather than a tidy conclusion, because I don’t have one.

Some days, running 20 AI agents across 20 terminal windows feels like I’ve finally found the operating system my brain was designed for. I build more in a morning than I used to in a week. The agency is stronger. The tools are better. The clients are happier. When neurodiverse teams are properly supported, they’re 30% more productive than neurotypical ones. I believe that, because I’m living it.

Other days, I close my laptop and my brain feels like it’s been microwaved. I’ve done extraordinary amounts of work, but I can’t remember eating lunch. I’m physically home but not mentally home. The boundary between “productive” and “compulsive” gets blurry when the tools are this good.

The honest answer is: AI agents have made my neurodivergent brain more capable than it’s ever been. They’ve also made it easier than ever to overdo it. Both of those things are true at the same time.

I don’t think the solution is to use less AI. I think the solution is the same boring thing it’s always been: self-awareness, boundaries, knowing when to close the laptop and actually be present. The tools don’t change the fundamental challenge of having a brain that doesn’t know when to stop. They just raise the stakes.

If you’re neurodivergent, or you suspect you might be, and you haven’t tried working with AI agents, I’d genuinely encourage it. Not the chatbot stuff. Proper agentic AI that holds context and works alongside you. Tools like Claude Code, Codex, Cowork. It might change how you think about your own brain.

Just maybe don’t open 20 terminal windows on day one.

Frequently asked questions