Don't burn the context

The big players have been laying off a lot of people lately — Amazon, Oracle, and a couple dozen others have cited AI in this year’s cuts — on the hopeful expectation that AI can do their work or make it unnecessary. Everyone watches the big players, so the question comes up: should we do that too?

Probably not. It is better to let your competitors try this radical maneuver first. There are a few reasons, but the biggest one is simple:

Context is the raw material

Context is the raw material AI reasons from — without it, the reasoning is shallow or suspect. Deprived of your context, the AI reasons from what it can infer about your operations from its training data — a woefully incomplete and probably inaccurate picture of your actual business.

Your marketing strategy is just that: your marketing strategy. For AI to reason well about it, it has to know what the strategy is and why it’s yours. The same goes for your customer support procedures, your IT processes, your products, your sales. You don’t want AI assuming that every function of your company is just like that of some hallucinated “average company.”

The Hansel maneuver

Your intrepid competitor — call them “Hansel, Inc.,” formerly “Hansel & Gretel Partners,” owned since 2023 by Grimm Capital — is in roughly the same place as your company, and indeed as most companies: its context is not very legible. Some context is splattered across these Word docs over here, or these PDF docs over there, or in Confluence, Jira, some Teams or Slack chat — you get the picture. Some important context is known only by a few people, or by only one person.

Hansel has in fact lost some important context about itself over the years, as knowledge walks out the door in the heads that hold it. Hansel, of all companies, should know how this goes: you mean to leave a trail behind you, and the birds eat it. So say Hansel lays off a significant fraction of its workforce. What then?

Hansel is not going to replace most of those people with AI overnight. In the meantime the survivors of the purge must shoulder their work; those beleaguered souls now have zero bandwidth. But the layoff frees up enough money to hire a bunch of very clever AI architects and engineers.

It is all very exciting at first. The architects whiteboard. The engineers prototype. Slide decks circulate.

Then the prototypes start to disappoint. The AI’s answers about Hansel’s actual operations are plausible-sounding but not quite right. The AI experts can deliver “their end,” but for good results they need significant time and effort from the people with the best grasp of context. Why does the Jira board work this way? Why are returns processed like this? How do we evaluate leads?

A lot of the knowledge needed lives in the heads of just a few people. It may be documented somewhere — but only those same people can say where, or whether the documentation is still true. Hansel needs them to help the AI team reconstruct context legibly, and it turns out they are the most overburdened people in the company. They always were! And now some of the heads that held the context are gone.

What does Hansel do now? Hire people back? (“Our bad, we forgot to pick your brain properly before.”) Hire a team of consultants to put it all back together? Decide to stop doing some good and important things, to free up the resources this requires? Throw up their hands and adopt some one-size-fits-all AI-for-business solution that isn’t right for their business?

These are not hypothetical questions. Klarna cut its way to an AI-first customer-service operation, then conceded the quality had suffered and began hiring humans again. Forrester finds that 55 percent of employers regret their AI-driven staff reductions, and predicts that half of AI-attributed layoffs will be quietly reversed. A Ford executive, explaining why the company was rehiring engineers, put the reason plainly: AI “is only as good as the information you use to train it.”

Hansel’s radical maneuver thrusts them into a world of unmet expectations, risks, and costs. Some of the risks may be severe. Say Hansel’s layoffs cut deep into IT and the engineers who maintain its software. One unfortunate aspect of the AI revolution is that the attackers got the new tools too: by IBM’s count, one in four breaches now involves AI, up by more than half in a year. Hansel still runs on software — software now defended by fewer people against better-armed adversaries.

You can empathize with Hansel; you can see how they got here. They saw correctly that the AI revolution is moving fast — the models, the APIs, and the cloud capacity are already there — and this seemed to call for swift and bold action. And they may yet be proven right that AI can eventually do much of what their laid-off employees did. But “eventually” is doing the work. The gap between the layoff and the capability is exactly when context bleeds out — and once it’s gone, you can’t get it back. Hansel didn’t just bet on AI; they destroyed the context that would have made the bet work. Swift, bold, and unwise. Oh well: “Never interrupt the adversary when they are making a mistake!”

Make the context legible instead

You absolutely should be thinking about what AI can do for your business. You should be thinking hard, and soon.

But the move is not to gut the team. The move is to make the context legible — to your people and to AI, in the same place — so that AI capability can compound on top of it instead of stalling out for lack of it. That work is unglamorous and it pays off slowly, but it’s the work that actually lets AI do useful reasoning about your business. Wiser to make smaller, targeted moves with legibility front and center — more on that, and on other aspects of strategy, in the pieces to come.

If you play your cards right, you may soon be looking in the rear-view mirror at a Hansel truck sitting jack-knifed in the middle of the highway.