Zwischen

The market cooled. Now you can hear yourself think.

A colder AI market takes the audience out of your AI decisions. That's the opening to plan what the technology is for, and what it means for your people.

7 min read

The midterms were Tuesday. On Thursday, the Nasdaq fell 5.5% and the semiconductor index about 12%, with Nvidia down about 9% and AMD about 17%. Reuters tied the rout partly to fears that the Democrats’ gains would slow data-center expansion, on top of standing worries about financing and returns (‘Investors wipe $2.5 trillion off Nasdaq as blue tsunami prompts AI fears’, Reuters, November 5, 2026). The same day, Reuters reported that prospective Anthropic investors were questioning $1.5 trillion ahead of next week’s roadshow, and some wanted a bigger discount (‘AI backlash could threaten Anthropic IPO timing, potential investors say’, Reuters, November 5, 2026).

By late Friday morning, the Nasdaq was up 1.7% (‘Live markets’, Reuters, November 6, 2026). Markets do that.

Here’s our guess, and it’s a guess: AI valuations are overcooked, and they cool from here. Anthropic may still pop on its first day. A pop wouldn’t answer the question we ended on in Anthropic booked the hall, which is whether the hall fills up.

We promised to take the other side of that question: what this means for the companies buying AI, not selling it. Our answer is more cheerful than the tape. A cooler market gives you your agenda back.

Who’s been setting your AI priorities

Most companies’ customers don’t buy AI. They buy insurance, or freight, or accounting software. AI still landed near the top of a lot of those companies’ plans, and not always for reasons of their own.

IBM surveyed 2,000 CEOs in early 2025. 64% said fear of falling behind had driven some technology investment before they clearly understood its value (‘IBM Study: CEOs Double Down on AI While Navigating Enterprise Hurdles’, IBM, May 6, 2025).

That’s the machinery behind Use more AI, they said. Capital markets got excited. Boards absorbed the excitement. Executives turned it into a mandate, and somewhere downstream everybody got a $200 Claude license and a usage dashboard. Speed became something you could show before anyone had settled the purpose.

THE AGENDA
Your customers didn’t write your AI priorities. Somebody’s investors did.

Our forecast is that this runs in reverse. The big platforms are already being rewarded for asking what the spending earns. That attitude travels to other boards, and from there to companies that never had a business reason to feel the pressure in the first place. Nobody has watched that cascade play out across the economy. We think it’s coming.

The buyers got permission first

The day before the rout, the market had already split the AI trade. Amazon, Meta and Alphabet rose while chipmakers and data-center operators fell, and Reuters read it as investors welcoming a more rational spending path from the companies paying for the buildout (‘Markets cheer Democratic win, betting divided Washington means fewer policy shocks’, Reuters, November 4, 2026). Investors cheered the companies buying the capacity and sold the companies supplying it. The buyer’s interests and the supplier’s growth story can come apart, and the market said so out loud.

A few days earlier, S&P moved Meta’s outlook to negative over capex and cash-flow uncertainty, leaving its AA- rating where it was (‘S&P Global revises Meta’s outlook to “negative” on AI spending, cash flow worries’, Reuters, October 30, 2026). Not a crisis. A signal that the bill is now a legitimate topic.

If Meta can be asked what its spending earns, so can your AI line item. The question moves from “are we doing enough AI?” to “what is this paying for?” That’s permission to kill the weak projects and keep the good ones, which beats defending all of them.

THE PERMISSION
The market just rewarded the buyers for asking what AI earns. You can ask too.

A license changes a task, not a company

The licenses weren’t worthless. A study of 7,137 workers across 66 firms found that regular Copilot users cut their email time by 3.6 hours a week, or 31%. Averaged across everyone offered the tool, the saving was 1.3 hours. Meeting time didn’t significantly change (‘Shifting Work Patterns with Generative AI’, NBER, May 2025).

The authors draw the line in the obvious place. People changed what they could change alone. They didn’t change what needed someone else to agree.

An employee with a license can draft faster. They can’t remove an approval step, change who sits in the meeting, merge two jobs across departments or decide the team stops producing the Thursday report. Those are management decisions. The blanket mandate asked individuals to find changes that only management had the authority to make.

THE LIMIT
A license changes what people can change alone. The rest takes a manager.

That’s what “people and process” actually means. Not change-management slides. The decisions and relationships around the work.

Who works with whom

At P&G, researchers put 791 professionals on real product-innovation problems. Individuals using AI matched the average performance of two-person teams without it. Commercial and R&D people working with AI produced proposals more balanced across both specialties, and teams with AI were roughly three times as likely as solo workers without it to land a top-decile idea. The effects were stronger on generating ideas than on choosing among them (‘The Cybernetic Teammate: A Field Experiment on Generative AI and Teamwork’, Organization Science, June 2026).

So the question isn’t how much more everyone can produce. It’s who needs to be in the room now that one person can carry part of another discipline’s knowledge, and where human judgment still decides what gets built. That’s a call about team composition and roles, and no license makes it for you.

AI changes who needs to be in the room. Somebody still has to decide who that is.

Plan it while it’s quiet

The mandate started with the tool and handed people a quota. Run it the other way: start with the business, end with the people, and let the tool land where it’s useful. None of this needs a transformation office.

Pick three things. Not an AI strategy. Three problems tied to what the business is trying to do this year: a margin you can’t hit, a customer you can’t afford to serve, a backlog that’s been the same length for two years. Make each case on your own economics. Your deployment doesn’t have to justify anyone’s $1.5 trillion valuation. It has to beat what it costs you.

Ask the people who do the work. They know where the time goes and which steps exist because someone, once, wanted a sign-off. They’re also the people who read “use more AI” as a performance review. Give them the problem, not the quota.

Redraw the job, not the task. For each of the three, write down what changes: which approvals go, who no longer needs to be in the meeting, which roles combine, what the team stops producing. If nothing on that list changes, you bought faster email.

Decide what the time is for before you find it. If a team gets a day a week back, say now what that day goes to. Otherwise it fills with more of what was already there, and the people who just got more efficient start wondering whether they’re next.

Measure the work, not the tool. Token counts and license activations tell you people used it. They don’t tell you anything got done. If marketing drafts twice as fast and legal reviews at the same speed, nothing ships sooner. The line just moved down the hall. Measure the finished thing.

THE TEST
If the finished thing didn’t get faster, nothing did.

Ahead of the next memo

The pressure doesn’t go away when the market cools. It changes its question. Last year the board asked whether you were doing enough AI. Next year it asks what the AI earned. The companies with an answer will be the ones that used the quiet to redesign a few jobs properly. Everyone else will be screenshotting a usage dashboard.

And the money will come back. When it does, it’ll bring a new mandate with it. You want the plan on the table before that memo lands.

The labs built the models. The market set the pace. Neither knows your business, and neither will redesign a single job in it. That was always going to be your people, and they’re the ones who’ll be living in whatever you decide.

The mandate was about the tool. The plan is about your people.

Further reading.

The answer is 14. Now what?

AI will build you 87 flawless charts before your coffee is cold. The old data hands knew the only question that mattered came after the number, and nobody is asking it anymore.

8 min read

All posts