Run ten things at once, but only look at three
Scott Batson got five times faster with AI, took on five times the work and burnt out. His fix is to slow down. Ours is a filter.
26 Oct 2026 · 4 min read
By Scott Batson on Blog
The short version
- Scott Batson, a staff engineer at Medium, found AI made him about five times as productive.
- So he took on five times as much: stacks of pull requests, tickets, threads and summaries, all at once.
- A few months in he was burnt out. Context got lost, work sat forgotten for weeks and his reviews got sloppier.
- His fix: while an agent works, don't start another one. Read properly, pair with a colleague or write.
- He now ships fewer things, but says they matter far more.
Who it's for: Anyone running several AI chats or agents at once, not just programmers.
Why it's genius
Most writing about AI and work is about doing more. This is an honest account of what happens when you actually do: a very experienced engineer, with all the right tools, wearing himself out in a few months.
His answer is refreshingly human. The gaps while an agent works are the best time for slow work: reading documentation properly instead of skimming, pairing with someone, writing up his thinking. He says that deeper reading made him better at steering his agents, because he now understands the whole shape of a feature before he hands it over.
It's also a funny, warm read. The point lands because he isn't preaching from the outside. He went all in first.
Try it
Write down everything on your list for today and start the ones AI can run on its own.
Before you switch away from anything, write one line on where you left it.
When results come back, pick the three that matter most. Deal with those first and let the rest wait.
While agents are working, do one piece of slow work instead of starting another task: read the docs, pair with someone or write.
When you finish, give me a three-line handover: what I asked for, what you did, and the one decision you need from me. If you need nothing from me, say so.
Our take
Parallel wasn't the problem. Nothing ranked it.
Read his symptoms again: lost context, forgotten pull requests, thinner reviews. That isn't what happens when you work in parallel. It's what happens when everything you start can land on you at once, with nothing deciding what matters most.
We think running lots of things at once is the right instinct. A to-do list used to be a queue: do one thing, cross it off, start the next. With AI it can be a launch pad. Fire off ten tasks, and let the agents work while you don't. The catch is the moment they come back. If all ten want you at the same time, you need to know which three matter most, finish those, and let the rest wait their turn.
Context switching is a systems problem
You can't train your way out of this. Psychologists have measured the cost of switching between tasks for decades, and practice shrinks it without ever removing it. Researcher Sophie Leroy calls the leftover drag “attention residue”: part of your mind stays on the task you just left.
The good news from the same research is that a short note on where you left off, written before you switch, reduces that residue. That's a system, not willpower. So the job is to make each switch cheap: every task that comes back should tell you what you asked, what was done and the one thing it needs from you.
Where we agree with him
Ten agents running doesn't mean you should be busy ten times over. The time they spend working is yours. Spend it on the slow work he describes, or on the three things that need you, and not on starting an eleventh task.
Where it shines
An honest, first-hand account of AI burnout from someone who went all in, with a practical answer for the waiting time.
Where it breaks
It's one person's story, and slowing down is only half an answer. Parallel AI work is here to stay; what's missing is a way to choose what gets your attention.
Where Ugenie helps
This is what Ugenie is for. Fire off as many tasks as you like, and Ugenie only brings you the three main tasks that need you next, each with a short note on where things stand.
Get Ugenie →Tools mentioned
Claude Code
Anthropic's coding agent. You give it a task in plain English and it reads, writes and runs code on your computer.
- Cost
- Comes with paid Claude plans, or pay as you go through the API.
- Our verdict
- The strongest coding agent we've used, and we build Ugenie with it. It isn't just for developers any more, but you do need to be comfortable with a terminal or the desktop app.
Coding agents in general
AI tools that can take a task, write and run code, and keep going until it's done. Claude Code, OpenAI's Codex and Google's Jules are the best known.
- Cost
- Most come with a paid plan from the AI company, from about £20 a month.
- Our verdict
- Worth trying even if you don't write code. The research found people in sales, legal and finance getting more done with them too.
Prices are rough and change often. Check before you buy.
Questions people ask
Is it better to run AI tasks in parallel or one at a time?
In parallel, as long as something decides what reaches you. Let the agents work at once, then deal with the three results that matter most before the rest.
Can you train yourself to context switch better?
A little. Practice reduces the cost of switching but never removes it. A short note on where you left off before each switch helps more than willpower.
What should I do while an AI agent is working?
Not start another task. Use the time for slow work, like reading something properly, pairing with a colleague or writing up your thinking.
Jargon in this piece
New to a word? Each one, in plain English.
All credit to Scott Batson. We share things we think are genius and add our own take. Watch or read the original for the full thing.