Building Systems for a Neurodivergent Brain

My brain doesn’t do reminders. Not the way they’re designed, anyway.

I’ll set a notification — “review weekly goals, 9am Monday” — and within two weeks it becomes wallpaper. The phone buzzes, I swipe it away before the words even register. It’s not that I forgot the thing. It’s that the reminder became invisible. My brain categorized it as noise and stopped surfacing it to conscious attention.

This is the central problem of ADHD executive function: the systems everyone else uses to compensate for being human don’t work for you. So you build your own.

The reminder problem

Here’s what I figured out about why standard reminders fail me. Two things need to be true simultaneously:

The reminder has to be invasive. Not a quiet notification badge. Not a tasteful little dot. It needs to interrupt whatever I’m doing, demand attention, be slightly annoying. If I can ignore it passively, I will — not by choice, but by neurology.
It has to be easy to set up, edit, and dismiss. Because if there’s friction in managing the reminder system, I won’t maintain it. I’ll set things up in a burst of motivation, then never update them when priorities shift, and within a month I’m ignoring a stale list of things that don’t matter anymore.

Most tools optimize for one or the other. Phone notifications are easy to manage but trivial to dismiss. Sticky notes on your monitor are invasive but impossible to maintain at scale. The intersection — invasive AND frictionless — is surprisingly empty.

And this matters more than almost anything else. Before a behavior becomes habit, before a mindset gets internalized, before a value becomes automatic — you need something external holding the pattern in place. A scaffolding for the thing you’re trying to build inside your head.

The brutal part: 80% of the time, I’m not even asking myself the right questions. Not because I don’t know the questions — I’ve written them down, I’ve thought about them deeply. I just… don’t think to ask them in the moment. The reminder system isn’t about memory. It’s about attention routing.

The capture funnel

The other half of the problem isn’t reminders going in — it’s ideas going out.

My brain generates ideas the way a popcorn machine pops corn. Constantly, unpredictably, and if you don’t catch them immediately, they’re gone. Or worse — they pile up. Uncaptured ideas create a mental traffic jam, this low-grade anxiety of “I had a thought and I lost it and it might have been important.”

So I built a capture funnel. The pipeline looks like this:

Idea hits → Dictate → Send to LLM → Get structured output

Voice capture is non-negotiable. When an idea strikes, I’m usually doing something else — commuting, showering, in the middle of a code review. Typing is too slow and requires too much context switching. Dictation lets me dump the raw thought in 10 seconds, messy and incomplete, and move on.

The LLM step is the key insight. Raw voice dumps are useless for later processing — they’re rambling, contextless, half in Chinese. An LLM cleans them up, extracts the core idea, and routes it to the right inbox. It turns a messy signal into something I can actually triage.

But here’s the part that took me longer to learn: you have to trust the system. The whole point of a capture funnel is that you stop holding ideas in your head. You let them go. The system holds them, surfaces the ones worth acting on, and lets the rest sit without guilt.

This is harder than it sounds. ADHD brains are wired to feel like every idea is urgent, every thought needs immediate action. Building a system you trust enough to not act on every impulse — that’s the real engineering challenge. The code is easy. The letting go is hard.

Aesthetic equals functional

I picked up this line from an ADHD organization video and it rewired how I think about tooling: every aesthetic choice should be functional for you.

Not “looks nice.” Functional. Reduces friction. Makes the right action easier than the wrong one.

This is why I use Emacs with Org-mode instead of Notion or Obsidian. Not because it’s objectively better — it’s objectively worse for most people. But for my brain, it works because:

It’s ugly enough to be honest. No beautiful dashboards that make you feel productive for opening the app. Plain text forces you to actually do the thing.
It’s keyboard-driven. No mouse, no menus, no clicking through UI. My hands stay on the keyboard, which means the gap between “I should capture this” and “I captured this” is about two seconds.
It’s programmable. When a workflow doesn’t fit, I change the tool. I don’t change my behavior to fit the tool. This matters enormously for neurodivergent brains — the tool adapts to you, not the other way around.

The same principle applies everywhere. My terminal prompt is minimal because visual noise costs attention. My file naming is aggressive — ONEPAGE_HEALTH.org, tracker_sleep.org — because I need to know what something is without opening it. My capture shortcuts are muscle memory because the moment I have to think about how to capture something, the thought is gone.

Every choice is a friction calculation. Will this make the right behavior easier? Keep it. Will this add a decision point? Kill it.

What I actually built

Here’s the concrete system, as of mid-2025.

The inbox: zinbox.org in Emacs Org-mode. Everything lands here first. Ideas, tasks, articles to read, random thoughts, meeting notes. It’s intentionally unstructured — the only rule is that things go in, and periodically I process them out.

Voice capture: Phone dictation → LLM processing → inbox entry. The LLM step converts rambling voice notes into structured items with tags and context. This is the “idea hits → dictate → send to LLM → get output” pipeline from my notes, and it actually works. The key is that capture takes under 15 seconds. Anything longer and I won’t do it.

The processing loop: Weekly (when I’m disciplined) or biweekly (when I’m honest), I sit down and process the inbox. Items get routed to project files, task lists, or the archive. Some get promoted to their own tracking files. Most get deleted — and that’s fine. The funnel is supposed to be lossy.

The reminder layer: Org-mode’s agenda system with aggressive scheduling. Deadlines show up in my daily view, and because Emacs is always open (it’s my entire work environment), I can’t avoid seeing them. This is the “invasive” part — the reminders live where I already am, not in a separate app I have to remember to check.

LLM-assisted processing: Claude Code as a thinking partner for processing complex inbox items. When I’m overwhelmed by a pile of captured ideas — and the feeling of overwhelm is itself an ADHD symptom, not just a workload problem — I can dump the whole mess into a conversation and get help triaging. “Here are 30 ideas I captured this week. Help me figure out which 3 actually matter.”

The meta-system: This whole setup lives in plain text files synced across devices. No vendor lock-in, no subscription, no app that might pivot to AI features I don’t want. Org files are just text. They’ll outlive every productivity app on the market.

The real lesson

I spent years thinking my productivity problem was discipline. If I just tried harder, focused more, built better habits. The usual advice.

Turns out the problem was architectural. My brain has different constraints than the systems were designed for. Standard reminders assume you’ll notice them — I won’t. Standard capture tools assume you’ll process later — I’ll forget. Standard organization assumes consistent executive function — mine fluctuates wildly.

So you build around it. You make reminders invasive. You make capture instant. You make the system hold the state your brain can’t. You accept that 80% of what you capture will be noise, and you build a funnel that makes the filtering cheap.

The system isn’t perfect. Some weeks the inbox piles up. Some weeks I don’t process anything. But the worst case is “stuff accumulates and waits” — not “stuff is lost forever.” That difference is everything.