════════════════════════════════════════════════════════════════════════════════
 Node 1  ·  2400 baud  ·  44 msgs in 6 areas  ·  sysop randogoth  ·  Mastodon
════════════════════════════════════════════════════════════════════════════════
════════════════════════════════════════════════════════════════════════════════
 Msg  : spoon-overdraft                              Area : GONZO
 From : randogoth                                    Date : 11 Nov 2026
 Subj : Spoon Overdraft Keys : gonzo, adhd, ai, english
 Msg  : spoon-overdraft
 Area : GONZO
 From : randogoth
 Date : 11 Nov 2026
 Subj : Spoon Overdraft
 Keys : gonzo, adhd, ai, english
════════════════════════════════════════════════════════════════════════════════

a report from the far side of a productivity binge. all observations biased, some exaggerated, the exhaustion is not.

It's November, and I'm sitting on the sofa, computer on my lap, a blank terminal open waiting for input while I stare out of the window, mezmerized by the leaves on the tree in front of it, moving in the wind. I can't remember what I wanted to do and I'm tired in a way that sleep doesn't fix. Behind me lies the most productive month of my life: thirteen projects moved, revived or finished, all my work tasks done, and even more private ones: a stack of devlog posts, three websites brought back from the dead or built from scratch, and a handful of decisions finally made that had been chewing on me for the better part of a year. I did a a whole year's work in a month. But to be honest, I want the month back.

It's funny though. The thing that got me here is the same thing that made my brain feel like it was working with me instead of against me for the first time for... like ever. How the hell can both be true?

Neurospicy Plumbing

I'm what is known as AuDHD, ADHD and some kind of 'tism, all in one package, all undiagnosed. If you share this combo you'll probably recognise some of what follows, probably with some discomfort.

Tell me a task needs to be done and you bet I dig my heels in, even when I'm the one who decided it needs doing, which is literally my job as a project manager. That reflex is called PDA which stands for "Persistent Drive for Autonomy", or more classical, "Pathological Demand Avoidance", and it doesn't care whose demand it is. A "blank page" (or terminal) can hold me hostage for hours and procrastinate for days. If my brain has found something more interesting to process, which it does most of the time (leaves! birds!), everything else turns grey. Or the exact opposite: intense hyperfocus so immersive I can't climb out of it, not even for food or for the thing that's actually due tomorrow. It's a superpower that defies control. Underneath it all runs a perfectionism that won't let me finish anything that isn't right yet, without any definition of what that would even be. And it applies to almost everything.

The worst are the loops. Decisions I can't make because every viable option clashes with some weird rule in my head. So the project stalls, then the next one that depends on it, and the loop keeps looping idle for months. Meanwhile I crave results like other people crave sugar, and I dread every chore that would actually put the loop into forward gear: testing, planning, proper documentation. I know they matter. They just don't give me fizzy feels.

Yes, you heard that right: I'm a project manager! How tf? The trick is that when it's someone else's project it usually goes smoothly. But my own projects? Oh boy!

Outsourced Executive Function

This year I started working with LLM agents properly (Claude, Mistral, Gwen), testing their abilities across different sections of software engineering and app development. I treat them mostly as collaborators and found out by accident that they work really well as something like an external "executive function", which is the thing that people like me struggle with.

PDA first: when a chore feels like a demand, and they all do, I hand it to an agent. Sometimes just with a prompt to break the chore down into manageable tasks and present a plan. This is already half of the cure, because once the thing has started, it isn't a blank page anymore but in motion that I'm part of, and the dread mostly evaporates. It becomes something we're solving together. Sounds silly, I know. But it works.

Even better on the loops. An LLM agent doesn't carry the rigid rules I carry around in my head. It is perfectly capable of laying out the options for how to proceed based solely on the data. It keeps all the threads in view at once and nudges me past my own emotional baggage, past the point where I'd normally circle back to the start because of some doubt or rule that got triggered. The oldest and heaviest of those knots was how to present myself online at all, and years of accumulated loops came undone in one long day. I already asked Claude to tell its side of that story, so I'll spare you the repeat.

Now to the chores: I write precise guardrails, strict rules for how I want code to be laid out, planned, written, tested, documented and committed, and the agent does the boring parts to my standards. My perfectionism finally gets fed instead of getting stuck. And because I ask the agent to track every tasks, subtask, and the time spent on them, I can finally allow myself to drop out of a hyperfocus more easily without that usual feeling of fear and guilt about abandoning it. Someone is still on it, keeping track, while I can go deal with the urgent thing.

The setup behind all this is a master AGENTS.md template I adapt for every project, time tracking in a shared todo.txt, and a skill that tells the agent how to talk to my ADHD brain. All of that probably deserves its own future post.

Ego Meets Expertise

I know how this sounds. There's a lot of hostility toward letting LLMs anywhere near a project, and most of what gets dismissed as "slop" deserves every bit of that criticism (except the term I try to avoid that has troubling roots in antisemitic internet slang). I'm by all means not asking anyone to buy into the LLM hype for all the legitimate reasons raised across the online debates. All I'm asking you is to trust that my brain has standards that I myself can barely meet, let alone an LLM. I don't hand anything over without enough detail, constraints, and a lot of overthinking, to meet them. The results are far from perfect, I gave up on that concept a long time ago, but they also ain't worse than if I'd do them by hand...

In fact results mostly exceed expectations. I've been writing software for decades and I'm proud of most of it, but my wiring has always cost my projects something, and I know it. When I let LLM review my code it tends to find bugs I couldn't, and comes up with improvements I hadn't thought of. One recent example that stung was orvanitag, a tool I had put a lot of thought and effort into. It clearly came back better than I'd left it and that in a fraction of the time I'd have spend on improving it.

That shit jabs at the ego, I'll admit. It's frustrating and often depressing, and I suspect that bruised ego is one of the bigger reasons AI is so hated in FOSS circles, even if nobody wants to openly admit it. In my own projects it turns out my hunger for quality is bigger than my ego, so my solution is: argue. I listen to what improvements LLM proposes and then I usually push back at it and refuse to implement a change until I'm convinced it beats what I had. The arguing is where I learn a lot. One month of this taught me more about my own craft than years of trial and error and digging through stackexchange.

Junkienomics

And it felt amazing. I could jump between multiple projects the very second boredom hit, knowing the work went on behind me. Things that had been stuck for months and years started moving. Knots that had been gnawing at me came undone. For a few weeks I was plainly happier than I'd been in a long time.

Here's what I missed: the friction I'd been fighting all my life was also my brake.

LLM took implementation chores off my shoulders, and I assumed the mental load went with it. But it didn't! I got to realize it just changed shape. Every agent running in the background is output to review, a proposal to argue with, a thread to hold. Every switch between projects is a context to rebuild in my head. The number of decisions per hour exploded far beyond anything I had ever done before, and every single one of them felt light and doable. Progress and flow.

The gratification never stopped. Every finished task was a hit, so I started another project. And another. The ADHD junkie brain that had always been starved of results was suddenly fed on demand, and it did what starved things do. It's addictive, it's satisfying, and it's an exhaustion I have no reference point for. There's a fancy word for it, technostress, which sounds way too clinical for what it feels like from the inside.

I'm still in the middle of it: complete burnout. I haven't found a fix, and I would be lying if I wrapped this article up with a neat list of rules.

What I can tell anyone with a brain like mine who's about to try this: the tool will remove many obstacles you've been fighting all your life, but some of those obstacles were protecting you. Deciding in advance how much you'll start might be a healthy strategy, because the part of you that decides in the moment is exactly the part that's been waiting its whole life for exactly this. I learned this the hard way. Perhaps I need to add strict break times to the master AGENTS.md file.

This article was outlined in a conversation with Claude, which sorted my exhausted brain dump into an outline. It seemed dishonest to write about working this way and then pretend I didn't. It saved me a few hours of editing that I can now use to rest more.

Links

--- RANDOGOTH BBS
 * Origin: randogoth.com (1:1/1)
────────────────────────────────────────────────────────────────────────────────
 [I]ndex   [K]eywords   [F]eed   Main [M]enu   

 Command: