The Assay · 18 November 2026

Will AI cut out the middleman in your industry? The work that survives is often yours

AI is taking over much of the work of passing things along between a customer and the result. What it leaves is the judgement someone has to stand behind, and in many owner-managed businesses that is the owner’s, so the work that protects the business from the squeeze is the same work that ties it to you.

Think about the enquiries that come in because people need you to find something out for them.

Which part fits. Which supplier can deliver by the end of the month. What the job will cost, roughly, before anyone comes out to look.

Some of those enquiries may stop. A customer who used to ring you can now ask an AI instead. It can read the makers’ data sheets, compare two stockists and give an answer in under a minute. If that happens, you won’t hear about it.

Or the customer rings anyway, with a figure the AI gave them, and asks you to match it.

What big tech’s October results show about AI and the middle

Vin Vashishta, who runs a data and AI consultancy, went through the large technology companies’ results at the end of October and found one pattern in almost all of them. Use is going up. The margin on each use is falling.

Alphabet said its search queries grew at the fastest rate in five years. In Vashishta’s reading, searches in Google’s AI Mode make much less money. Reddit told investors its traffic from search was down “meaningfully” in the quarter. Vashishta puts that down to people reading an AI’s answer instead of clicking through.

In his reading, an AI agent collapses several steps into one screen, and he says agents do this “to every workflow they touch”.

For a smaller business it could look like this, as my example rather than his: where a customer used to search, read a handful of websites, ring a supplier and then buy, an agent can do it in one go.

Not every business in the middle loses out. In his words, they “all add cost, but not all of them add value efficiently.”

Those are results from some of the largest technology companies, read by one adviser. Whether and how fast the pattern reaches a UK business your size is not known. The direction is worth taking seriously.

Passing work along, and making a call

From his figures I draw a distinction he does not make.

Most businesses in the middle of a chain do two kinds of work, and they are easy to confuse.

The first is passing things along: finding the supplier, comparing the prices, chasing the delivery, reformatting one company’s information for another. Customers paid for this because doing it themselves took time. An AI can now do much of it in seconds, often at no cost to the person asking. That is the work being squeezed.

The second is making a call. Which of three quotes is safe to accept. Whether the building will pass its fire inspection with what is installed now. Which job to turn down because the customer will never pay on time.

Customers pay for this because getting it wrong costs them more than your fee. An AI can gather the information for it. It cannot carry the responsibility for it.

When I managed discretionary portfolios at a Swiss private bank, the decision on what each client held was mine, and so was answering for it.

The line moves. Some of what counts as a call today will become passing work, and the cheaper a mistake is, the sooner it moves.

The work that survives often sits with the owner

The second kind of work, the calls, is the work that survives. In many owner-managed businesses it is also the work that sits with the owner. The supplier you trust when the specification is unclear. Whether the old installation can stay or has to come out. Which customer gets the last of the stock.

Think about what happens to the business if the owner steps back for a month. The passing-along work carries on, and that is the work AI is taking over. The calls wait for you.

That puts a business in an odd position. It is protected from the squeeze by the same thing that makes it fragile.

Sort ten things your customers pay you for

This takes half an hour with a pad of paper.

Write down ten things your customers pay you for. Not your services as they appear on the website. What they actually rely on you for, in their words if you can. If you can’t think of ten, take the last ten enquiries the office dealt with and ask what each customer was paying you for.

Next to each, write one of two words.

Passing if a well-informed customer with an AI could do it themselves from public information in an afternoon.

Call if it needs a judgement that someone has to stand behind.

If a line is both, split it: the part anyone could look up, and the part someone has to stand behind. If most of what you sell is hands-on work, a qualified person on site, leave the work itself off the list. AI isn’t doing that either. List the questions and decisions around it.

Then, next to every Call, write the names of the people who can make it today without asking you. That is the same question as the budget test in The Assay No17, asked from the other end.

Three kinds of line come out of this:

  1. Passing lines are the work being squeezed. Use AI on them yourself, so the saving shows up in your margin first. One way to start: give it your standard price list and spec sheets and see whether it answers your five most common enquiries the way you would.
  2. Call lines with two or more names are the core of the business, and they do not depend on one person.
  3. Call lines where the only name is yours are where the business is both protected and fragile.

Then put a rough share of turnover next to each line. Add up the Passing lines: that is the part of your income most exposed to the squeeze.

Add up the Call lines with only your name: that is the part that will hold, for as long as you are there to hold it. Put the two numbers side by side.

My guess is that fewer of your Call lines have a second name than you expected. If the sort surprised you, the line that did it is the one worth sending me.

Sources: Vin Vashishta, “Did The AI Bubble Just Pop?”, Substack, 1 November 2026, https://vinvashishta.substack.com/p/did-the-ai-bubble-just-pop ; Alphabet, Q3 2026 results release (Form 8-K, Exhibit 99.1), https://www.sec.gov/Archives/edgar/data/0001652044/000165204426000080/googexhibit991q32026.htm ; Reddit, Q3 2026 Letter to Shareholders, https://s203.q4cdn.com/380862485/files/doc_financials/2026/q3/v2/Q3-26-Shareholder-Letter.pdf

Previously in The Assay: Who is responsible for what your chatbot says? What the Air Canada case shows

The Assay is written by Rudi Prenzlin at Scalable Growth, a UK practice that gives owner-managed businesses a fractional CFO and the operational structure to run without every decision going through the owner.

← All notes Begin a conversation