Signal Over Noise Vol 2, Issue 24 | Jun 19 2026**
**AI hasn’t made me faster — it’s made me able to start
Don’t have time to read this week’s issue? Why not copy/paste it into your AI agent and ask it for insights?
Last Tuesday I came out of a meeting with a narrow window before the school run, and inside of it I did the following:
- sent a client invoice and caught a tax error in it that a human would’ve waved through
- untangled a broken municipal payment portal
- chased a mismatched name on my kid’s foreign identity paperwork
- migrated a client off a fragile setup onto a hardened server
- confirmed my Spanish tax return with my accountant.
Then that evening, after the kids were down, I did my comms triage (mail, WhatsApp, Telegram) and looked at analytics across my sites and newsletter.
None of those tasks have anything in common. If anything, jumping between things this different is what wrecks my day — each one is a cold start, and the starting is the part that costs me. That Tuesday, it barely cost me anything, and that’s the change that using AI has made to my work.
The executive-function tax
I have ADHD (I don’t mind talking about it, it’s a late diagnosis within the last 5 years, so this is still relatively new ground for me). So when people say AI makes you “10× faster,” it lands wrong for me, because speed has never been my problem.
My problem is the gap between knowing I have to sit down and do a thing, and actually having the will and the drive to sit down and do it. That gap is called task initiation, and it’s one of the executive functions — the same family as planning, prioritising, and holding things in your head.
It’s a real cognitive skill, not a character flaw, and in an ADHD brain it’s specifically impaired. The same research describes the rest of my daily life with uncomfortable accuracy: weak prioritisation, time blindness, and “an inability to orchestrate concurrent tasks.”
Starting is hard, switching is hard, and holding it all at once is hard.
There’s a chemical floor under all of it. ADHD brains run short on dopamine in the reward pathway — a PET study from Volkow and colleagues found measurably decreased function in exactly that circuit, and tied it to the motivation deficit. In plain terms: if a task doesn’t produce a hit, it doesn’t get done (or it gets done so slowly and so late that it may as well not have).
So the simple, boring things with no spark slide to the bottom of the pile and stay there.
Look, even if you don’t have ADHD, you still know this feeling. You’ve stared at an email you could write in four minutes and not started it for two days. For me that’s not the exception — it’s most of the list, most of the time.
But the switching stopped costing me
Over the last couple of years, doing “actual work” itself didn’t get faster. What’s changed is that the switching has stopped costing me anything.
On a normal day, moving from a security problem to a tax form to a parenting errand means three separate cold starts, each one a small wall I have to climb.
The research calls this the switch cost — every context change makes your brain reload working memory and reset.
With an agent in the loop, I don’t pay that attention-span tax, because I’m not the one holding the context. The system holds it and I just bring the judgement.
A few concrete things it does for me:
- It’s a second pair of eyes that doesn’t get bored. Earlier this week a client invoice went out with a tax error in it (the kind Spanish tax rules make genuinely easy to miss), that a tired brain at the end of a long day would have waved through. It got caught. That’s a check I can’t reliably run on myself, and it doesn’t get bored of running it for me.
- It gets things out of my head. I don’t have to keep that paperwork deadline, the tax return, the broken portal in my main memory anymore. I can (literally, in voice mode) say: this is the issue, this is the due date, remind me when you think I need it. Working memory is the deficit I feel most, and this just routes around it.
- It does the triage I can’t. I had a stack of things due by Friday and no native ability to rank them. Handing that off — “what should I actually do first, in the next twenty minutes” — is the difference between a productive window and a paralysed one. I kick off a Claude session by saying: “Right, what’s next?”
The mechanism underneath is simple: I get a small dopamine hit off talking to the agent and watching the thing get done.
For most people that’s a nice-to-have; but for a brain short on reward chemistry, it’s the fuel that gets the boring task over the line at all.
Knowing where to put the stops
“AI fixed my ADHD” would be a lie, and the whole point of this newsletter is Signal Over Noise.
It doesn’t fix it. It makes mistakes, constantly, and you have to be willing to sit down and talk them through — and it gets genuinely frustrating when you’re correcting the same thing for the third time because it forgot what you told it. (That’s why I keep an external memory — Obsidian, a local search index — so the context survives between sessions and I’m making the judgement calls.)
But the real catch is subtler, and it’s the same property that helps. It’s an everything box. It can do anything, which means it will go anywhere you point it.
If you’ve got a busy brain that won’t be quiet — an idea about an idea about an idea — the agent will happily let you chase every one of them down.
The exact friction that used to stop me starting also used to stop me spinning out. Remove it, and you can spin out at machine speed.
So the skill that matters now is knowing where to put the stops. Saying: no, not that path right now — make a plan for it, flag it for next month, park it. The tool will enable you as far as you let it.
Which means you have to know yourself well enough to decide how far that is.
If your brain works like mine — or not
If your brain works like mine, the thing AI can change for you probably isn’t speed — it’s getting started at all. The cost of starting, switching, and holding too much at once is exactly what an agent can absorb, and that’s worth more than any benchmark.
And if your brain doesn’t work like mine, the same thing is true in a smaller way. Everyone has the four-minute email they can’t start. The lesson holds either way, and so does the caveat: it’s an everything box, it’ll enable your worst habits as readily as your best ones, and the judgement — about what’s worth doing, what’s true, and when to stop — is still yours.
Calling it a prosthesis (I think) feels about right: it doesn’t fix what’s missing, it gives me a way to work around it.
And for some of us, that’s the more useful thing to have.
Until next time,
Jim
The tools I built for exactly this
A few of these are open-source. I built them for my own brain; they’re free if yours works the same way. Just point your AI agent at the links below and ask what it can do with them.
- daily-patterns-pack — logs each session to my notes so tomorrow-me doesn’t have to remember, then flags what’s worth automating.
- lessons-learned — turns a mistake into a rule the system catches next time, so I don’t have to.
- think-first — makes me weigh a decision before I impulsively build it.
Second Brain Chronicles — what I built this week
The main thing I built this week was the call to retire something I’d already built.
I went into the little Kit MCP server I’d made a while back — planning to give it a security pass and add a feature or two — and found that Kit had quietly shipped their own official one. So before touching mine, I tested theirs. It’s more full-featured (roughly 70 tools to my 29), and it’s built and maintained by the Kit team. That pretty much cinched it.
Instead of extending my server, I deprecated it. I didn’t delete it — it’s still on GitHub, and I’m glad it existed and got used. I marked it deprecated, rewrote the README to point at the official one, and explained why.
I want to be straight about it: this wasn’t some noble open-source gesture. My server was never going to make money. It was a creative thing that proved there was demand for using AI agents to interface with the Kit mailing list SaaS — and then the people who own the API shipped the proper version. That was always going to happen, and it’s a good outcome, not a defeat.
If you’ve built a thin wrapper around someone else’s service — an MCP, a CLI, a small integration — that’s worth doing. It proves the need and it scratches your own itch. But hold it loosely. When the official version lands, the move isn’t to defend your patch of ground; it’s to point your handful of users at the better-supported thing and tell them why. Knowing when your tool’s job is done is its own small skill.
→ The full write-up, with the side-by-side tool breakdown, is on the Member Hub.
You’re reading the free edition of Signal Over Noise. If you liked this piece, you should sign up for the paid membership — €7 a month or €69 a year.
In addition to the newsletter, paid members get:
- No ads
- The Second Brain Chronicles — my “what I’ve actually been doing with AI this week” build log, inside every Friday issue
- Exclusive access to the Paid Members Notion Hub — every build, script, prompt and guide behind the newsletter, yours to copy