You've Always Given AI One Job at a Time. This Year, People Are Quietly Learning to Run a Small Team of Them
AI still mostly gets used one task at a time — ask, wait, check, repeat. That's quietly changing: people are starting to run small teams of AI helpers instead of one assistant. Here's why the old habit of checking the work matters more now, not less.
In this article
Most people still use AI the way you'd use a very fast intern: you give it one thing to do, you wait, it hands it back, you check it, and then you give it the next thing. One job, one turn, one at a time. That's been true since this whole wave of AI tools started, and honestly it still describes how most of us use it today.
But quietly, in the last few months, something has started to shift for the people building things with these tools — and it's not really about the AI getting smarter. It's about people learning to stop treating it like one assistant and starting to treat it like a small team.
The number that stopped me
Here's the detail that made me sit up. Engineers who use AI tools heavily now say AI touches something like six out of every ten hours of their working day. That's a huge number. You'd expect, from a number like that, that most of the actual work is basically being handed over.
It isn't. When you ask those same people how much of their work they'd trust to fully hand off — start to finish, no check-in, no second look — the honest answer is a small slice. Not most of it. A sliver.
Sit with that for a second, because it's the whole story in one line: people are using AI for almost everything, and trusting it completely with almost none of it. That gap between "touches everything" and "can be left alone" is exactly where the interesting change is happening.
What's actually changing
For the last couple of years, the pattern was: ask AI to do one thing, wait for it, look at what came back, then ask for the next thing. One helper, one task, your full attention required at every single step.
What's starting to happen now is that people are setting a few of these helpers loose on different pieces of the same job, all at once — one drafting something, another checking it against the facts, another formatting it or tidying it up — and only getting pulled in themselves when one of them hits a decision that genuinely needs a human. Some of them now keep working for hours, occasionally longer, checking in only at the points that actually matter, instead of after every tiny step.
Think about the difference between doing every task yourself and running a small, capable team. A good manager doesn't do everyone's job for them, and they don't walk away and hope for the best either. They decide what gets delegated, they set a clear idea of what "done well" looks like before the work starts, and they know exactly which moments deserve their own eyes before anything goes out the door. That's the shift happening right now, except the "team" is a handful of AI helpers instead of people, and a lot of us are learning this management skill for the first time without ever having called ourselves managers.
Why this raises the stakes on something I already warned about
I wrote a few weeks back about the gap between building something fast with AI and building something that actually holds up — that speed and durability are two different skills, and the businesses that skip the second one usually end up paying for it later, quietly, a few months down the line.
This new shift makes that same lesson matter more, not less. When it was one AI helper doing one task, there was one thing to check before you moved on. Now that people are setting several helpers loose on different parts of a job at the same time, there are several places something can quietly go sideways at once, all while you're only half paying attention because it "looks" like it's handling itself. More hands on the job doesn't mean less need for someone to check the work. If anything, it means you need to be clearer than ever about who's checking what, and when.
What this means for you, plainly
You don't need to run a tech company for this to matter to you. If you use any AI tool for your work — planning a project, drafting proposals, organizing your business's records, anything — the real skill worth building this year isn't "get better at prompting." It's the much older skill of being a good delegator: deciding clearly what you're willing to hand off, saying out loud (even just to yourself) what "done right" looks like before the work starts, and knowing in advance exactly which parts you personally need to look at before they go anywhere near a customer, a payment, or anyone's personal information.
That's a life skill, not just a tech one. The people who get good at directing a few capable helpers — human or otherwise — without either micromanaging every step or checking out completely, are going to have a real edge over the next few years. Everyone else is either going to burn out trying to do everything themselves, or get burned by trusting something they never actually checked.
A small habit for this year
Before you hand a piece of work to AI — especially if you're handing off more than one piece at a time — write one short sentence answering two questions: what does "done right" look like here, and who is the last human who looks at this before it goes out?
If the honest answer to the second question is "nobody," that's worth noticing before you move on, not after something breaks.
Where I land on this
I don't think this makes AI riskier than it already was — I think it just makes the old advice more urgent. The tools are getting genuinely capable of handling more at once, and that's a real gift if you're running a small business or a small team and you've never had the budget to hire the help you actually need. I use several of these tools myself, often at the same time, on different pieces of the same piece of work.
But capability was never really the hard part. Knowing what you're comfortable handing over, and what still deserves your own two eyes before it matters — that was the actual skill before any of this started, and it still is now. The people who do well with this next wave won't be the ones who hand off the most. They'll be the ones who always knew exactly which parts they weren't allowed to.
Steve Nyanumba
Building software for Kenyan and African businesses at Vapor Technologies.
Building something for your business?
We help Kenyan and African businesses ship software that performs.