I'm going to be honest about something that most people in tech don't talk about openly.
I have ADHD. I'm 40. I'm a self-taught developer with no CS degree. And I lose context constantly.
Not in the "oh I forgot where I put my keys" way. In the "I just spent 3 hours deep in a codebase, got interrupted by a Slack message, and now I genuinely cannot remember what I was doing or why" way.
If you have ADHD, you know exactly what I'm talking about. If you don't — imagine your browser crashing and losing 47 tabs. That panic? That's every Tuesday for me.
The AI Paradox for ADHD Brains
Here's the thing nobody warned me about: AI coding tools made my ADHD worse.
Not better. Worse.
Why? Because they're stateless. Every session is a blank slate. And for someone who already struggles with working memory, having to rebuild context from scratch every time I open Claude Code or ChatGPT is genuinely exhausting.
Me at 9am: "I'm working on a Next.js app with Supabase auth,
JWT refresh tokens, the user table has these columns..."
Me at 2pm: "I'm working on a Next.js app with Supabase auth,
JWT refresh tokens, the user table has these columns..."
Me at 9pm: "I'm working on a Next.js app with Supabase auth..."
I type the same context paragraph 5-8 times a day. Every. Single. Day.
For a neurotypical developer, this is annoying. For me, each repetition drains the limited executive function I have. By the third time, I'm frustrated. By the fifth, I'm done for the day.
The tool that's supposed to help me think is making me spend my mental energy on remembering what to tell it.
The Breaking Point
It happened on a Thursday at midnight. I was debugging a race condition. I'd been at it for two hours. Claude gave me a suggestion I'd already tried — because it didn't know I'd already tried it. Because it doesn't remember anything.
I typed (and I'm not proud of this):
"I ALREADY TRIED THAT. WE DISCUSSED THIS 20 MINUTES AGO IN THE PREVIOUS SESSION."
Then I realized: I was yelling at a machine for not having memory. And the machine was right to not remember — it was designed that way. Every AI tool is designed that way.
That's when something clicked. Not the bug (that took another hour). Something else:
The problem isn't AI intelligence. It's AI amnesia.
What I Actually Needed
I sat down and wrote a list of what would actually help my ADHD brain:
- Don't make me repeat context. Ever. If I told you once, you should know it forever.
- Remember what worked AND what didn't. So you don't suggest the thing I already tried and rejected.
- Know when I'm frustrated. If I've been debugging the same thing for 2 hours at midnight, maybe don't give me a 500-word explanation. Just give me the fix.
- Carry context across projects. My coding style doesn't change when I switch repos. My preferences don't reset.
- Forget things naturally. Not everything is worth remembering forever. That random CSS hack from 6 months ago? Let it fade.
I looked for tools that did this. I found vector databases pretending to be memory. I found RAG pipelines that retrieve documents but don't understand context. I found "memory" features that are just conversation logs with a search bar.
Nothing actually modeled how memory works. How my brain works (when it works).
So I Started Building
I'm not going to pitch you anything. But I will tell you what I learned, because maybe it's useful to someone else.
I started reading neuroscience papers. Real ones. Not Medium articles — actual research on how human memory works:
The Ebbinghaus forgetting curve — memories decay exponentially unless reinforced. Your brain doesn't delete things; they just fade. This is the opposite of how databases work (store forever, delete explicitly).
The PAD emotional model — Pleasure, Arousal, Dominance. Psychologists use three dimensions to map any emotional state. Turns out, emotions are not decorations on memory — they're part of the retrieval mechanism. You remember emotional events better. Your brain literally prioritizes them.
Circadian rhythms affect cognition — you're not the same developer at 9am and 11pm. Your ability to focus, recall, and make decisions fluctuates. Any system that treats you the same at all times is ignoring biology.
I started modeling these concepts in code. Not as a product — as a personal tool. Something that would sit between me and my AI tools and just... remember things for me.
The first version was ugly. 200 lines of JavaScript and a JSON file. But it worked. I told it my preferences once, and the next day, they were still there. I told it about a debugging approach that failed, and it remembered that too.
For the first time in years, I felt like my tools were adapting to me instead of the other way around.
What I Learned About Memory (and ADHD)
Building this taught me things about my own brain:
1. Context is not content. There's a difference between "what was said" and "what matters." Most AI memory solutions store conversations. But what I need is context — the decisions, the preferences, the patterns. Not the 400 lines of chat where we discussed them.
2. Forgetting is a feature, not a bug. My ADHD brain forgets things constantly, and yes, it's frustrating. But it also means I don't carry stale assumptions. I approach problems fresh. The best memory system isn't one that remembers everything — it's one that remembers the right things and lets the rest fade.
3. Emotions are metadata. When I'm frustrated with a solution, that frustration is information. It means "this approach has a problem, even if the code technically works." Any memory system that strips emotional context is throwing away signal.
4. My ADHD is a design constraint, not a disability. When I design for my own limitations — short working memory, need for external structure, sensitivity to context switches — I end up building things that work better for everyone. Turns out, neurotypical developers also hate re-explaining context to their AI tools. They're just more patient about it.
The Uncomfortable Truth
Here's what I think about late at night:
Every major AI company is racing to build bigger models, longer context windows, better reasoning. And that's great. But none of them are building memory.
Not real memory. Not the kind that persists, decays, carries emotion, and adapts to you over time.
They're building incredibly smart assistants with permanent amnesia. And we've all just... accepted that?
I didn't accept it. I couldn't. My brain wouldn't let me.
Where I Am Now
I'm a solo founder. Venezuelan, living in Medellín, Colombia. Zero employees. I run the company with AI agents (which is a whole other post).
I've been working on this problem for months. Some days I feel like I'm building something important. Other days I feel like I'm a guy with ADHD who got hyperfocused on a niche problem and can't stop.
Both are probably true.
If you're a developer with ADHD — or honestly, any developer who's tired of being the memory system for your AI tools — I'd love to hear how you deal with it. What workarounds have you found? What do you wish your tools remembered?
And if you're building AI tools: please, for the love of everything, add persistent memory. Your ADHD users will thank you. Your neurotypical users will thank you too, they just won't know why.
This is my first post here. I'm Mario. I'll probably write about building things alone, neurodivergent entrepreneurship, and why AI tools should be designed for how brains actually work — not how we wish they worked.
Top comments (0)