CAREER STRATEGY NEWSLETTER
We’re all used to self-checkouts in stores nowadays.
Think about the last time you used one.
The scanning is fast. Beep, beep, beep and in the bag. Nobody needs to watch you do it, because the machine already knows whether the barcode matched.
Then you put a bottle of wine through. A light flashes and everything stops.
Now you're waiting for the one person in a fleece who looks after the whole bank of tills. They're checking someone's ID. Then they're sorting out an unexpected item in the bagging area. Then a customer who's been charged twice and isn't happy about it.
The scanning got quicker. The queue moved to the one part of the process that still needs a person.
And where there used to be someone on every till, there's now one person for all of them.
Research published this week found the same pattern in highly qualified knowledge work. It says a lot about where senior careers are heading.
Let's dive in.
Where the time goes when AI is good at the work
On 5 November, one of the large AI labs published a study of 207,049 real research sessions on its AI tools, recorded between 2 August and 30 September 2026.
The question was simple. What do scientists actually use AI for?
The answer is lopsided:
Data analysis: 33.7% of sessions, against 27.6% of the time researchers normally spend on it.
Writing and communication: 32.0% of sessions, against 16.0% of researcher time.
Generating ideas and hypotheses: 4.2% of sessions, against 7.1% of researcher time.
The researchers also surveyed 528 scientists. Asked which work matters most for producing important results, they ranked generating ideas first. That's the area they use AI for least.
Then the researchers scored every research task on how well AI can do it. The tasks AI scored best on got about twice as much use as the time spent on them would predict.
Those tasks had a lot in common. Mostly thinking rather than physical. Done in text and numbers. Easy to check. Physical, interpersonal and regulated work all scored lower.
Here's the finding that stuck with me:
With today's AI, widely adopted, the researchers estimate it could take on 27% to 52% of the researcher time needed to produce a new study. A median of 38%.
Output rises far less than that. Between 1.35x in the arts and 2.02x in statistics.
Their explanation is one line: "Freed-up time does not translate one-for-one into more output." Researchers move their effort onto the tasks where AI helps less. The bottleneck moves with them.
(AI lab research, "AI for scientific discovery", 5 November 2026)
The caveats matter. These are scientists, not Directors. The survey covered active AI users, mainly in life sciences. The output figures come from a model, and the researchers call them scenarios rather than forecasts.
Then again, about 40% of the scientists they surveyed expect AI to carry out full research projects on its own by 2030. About 64% by 2035.
Faster scanning. Same queue. Different place.
Why "AI makes me faster" is only half the story
When I talk to senior people about research like this, most hear good news. AI takes the heavy lifting. They get time back. They become more productive.
I'd read it more carefully, for 2 reasons:
The first is what the study sorted on. Thinking work. Done in text and numbers. Easy to check. Put your own week through that filter…:
The forecasting and commentary that goes with it. The board paper. The market summary. The pricing model. The quarterly review deck. Most of what fills a Director's diary outside of calls is analysis and writing, and most of it can be checked in minutes. The numbers reconcile. The document reads well. The deck tells a clear story.
That's the scanning. It's also the work most of us were promoted for.
The second reason is what works against the productivity gain point of view. When a big share of the time is freed up, effort piles up at the steps AI can't do. The bottleneck becomes the job.
In research, the bottleneck is the physical lab work, and choosing which question is worth asking.
In a corporate role, I'd argue it looks like this: deciding which problem the business should solve. Holding on to a difficult customer. Conversation that needs someone with authority to make a cahnge or unblock a situation.
That's the person in the fleece.
And there's a catch! A shop that needs one person per bank of tills doesn't keep 6 people. The value concentrates at the bottleneck. The headcount doesn't follow it.
So seniority built mostly on high-quality analysis and writing is more exposed than it feels. Seniority built on choosing the question and owning the outcome is what survives the redesign. There are just fewer of those seats...
Here's why I'm telling you this in a career newsletter
Back to the supermarket.
Nobody trains to be the self-checkout assistant. The job appeared once scanningno longer needed a staff member. People who do it well were already good at the awkward bits that come with the job such as the judgement calls, dealing with upset customers, knowledge as to when to wave something through.
Your second option works the same way. A board seat, a fractional role or an advisory client is paying for the bottleneck: which question to ask, and the decisions that can't be checked yet. The analysis comes with it, or from a tool.
Most of us have never measured how much of our week is in active decision taking mode.
So run the checkout test. This week:
1) Sort last week by how it would be checked. For each block of work, ask whether a capable person could check it in under 10 minutes. A reconciled forecast, a summary, a deck: yes. That's scanning. Which market to enter, which hire to make, which account to protect, a hard conversation: no. That's the bottleneck.
2) Count the bottleneck hours. If it's a small share, that's normal. It's also the share likely to define your role in 3 years, whether you plan for it or not.
3) Test your questions outside your company. Pick a business you know from the outside. A supplier, a customer, a friend's company. Write the 3 questions you'd ask first if you joined their board tomorrow. Then ask someone there whether they're the right questions. If they are, that's a skill that travels.
None of this means using AI less. I use it every day. It means knowing which part of your work is the scanning, and building your options on the part that isn't, while you're still employed and the timing is yours.
This is the work Zelova is built for. The guide looks at your actual role, separates the work that's easy to check from the work that holds the queue, and works out who else would pay for the second kind. Then it turns that into a plan with more than one outcome in it. Built while you're employed, so the choice of terms stays with you.
It's a career strategy service, still being built. If any of this feels close to home, the waitlist is open at zelova.ai.
One last thing
Where does the queue form in your week? What's the work that waits for you, because nobody else can make the call?
Reply and tell me. I'd like to hear your version.
David
You built the audience. Can you reach them?
You can have thousands of people following your work and still need an algorithm’s permission to reach them. Creators know the drill: you post, you pray, you check the numbers, you try again tomorrow.
A newsletter changes that relationship. On beehiiv, you can build a subscriber list you own, reach readers directly, and turn their attention into a sustainable business with subscriptions, sponsors, a global ad network, and more.
A deal has two sides. Your stack only sees one.
Aligned shows you the other half. See what buyers actually do between meeting. See what they open, revisit, and share; so you know the exact next step to move things forward. And your champion gets everything they need to win internally when you're not in the room.
Why are so many brands shifting ad spend to CTV?
For marketers who want a presence in streaming, Roku is the obvious place to go. With 47% of all streaming time in the US, it has more reach than every competitor. Discover how Roku Ads Manager can spread your message across 100+ premium apps in a single, self-serve ad buy.



