We killed Jeeves in 2006. Then spent twenty years building him back.

On May 1st this year, Ask.com closed for good. The farewell message on their homepage read: "Every great search must come to an end." It signed off with five words that deserve some attention.

Jeeves' spirit endures.

Wind it back thirty years. Ask Jeeves launched in 1996 with a simple premise. Don't learn how to search. Just ask the butler. He'll fetch what you need. Type your question in plain English and Jeeves, impeccably dressed and unfailingly helpful, would trot off and return with an answer.

It was charming. It was intuitive. It was completely the wrong idea about what intelligence was for. By 2006, Jeeves was gone. Retired. The brand dropped him because he had become a liability. A metaphor that told users to expect servitude rather than search. Google had already shown that the right interface was not a butler. It was a blank box and a billion signals. Less personality. More power.

So Ask Jeeves became Ask and quietly faded out. We learned our lesson. Or so we thought.

We have spent the last four years building the most capable reasoning systems in human history. Models that can synthesise research, interrogate assumptions, write code, identify strategic risk, reframe problems and think across disciplines in ways that no single human brain can do alone. The dominant use case in many small businesses, right now, is: "write me a summary of this email."

I hear this constantly. Across sectors, across company sizes, from people who should know better. "Butler-ism". The reduction of extraordinary intelligence to domestic tasks. The new Jeeves, wearing a different coat. The entry point makes sense. AI is easiest to adopt at the task level. The stuff you were already doing, slightly automated. Summarise. Draft. Reformat. Translate. It feels like efficiency. It produces quick wins. It justifies the licence fee.

But it is a misreading of what the moment requires.

Companies treating AI as a production accelerant will get exactly that. More output. Faster. Cheaper. In a world where everyone has access to the same accelerant, faster and cheaper is not a competitive position. It is a commodity race with a predictable winner. Whoever spends the most on compute.

Companies treating AI as a thinking partner, putting it upstream of execution, inside strategy, at the point where problems are still being defined, are the ones building something that compounds. There is a word for the difference. Jensen Huang used it recently, talking about software engineers, though it applies just as well here.

Purpose versus task.

If your purpose is to execute tasks someone else has defined, you will be replaced by a system that executes them cheaper and faster. That is not a threat. It is a description of what is already happening.

If your purpose is to find the right problems, to interrogate the brief before you answer it, to bring genuine judgement to what is worth doing and why, that is something different. That is not Jeeves. That is the person who employed Jeeves.

While all of this is happening, something else is happening too. And the gap between the two is widening by the day.

Upstairs, a different world entirely.

The frontier organisations are deploying AI that plans, codes, reasons and executes, with minimal human involvement. Autonomous agents that run entire workflows end to end. Systems that do not wait to be asked. They identify the problem, devise the approach, build the solution and deliver it. Gartner calls this the most aggressive adoption curve of any emerging technology it has ever measured. The organisations building here are not thinking about email summaries. They have moved past the question of what AI can do for them. They are asking what becomes possible when human execution is no longer the constraint.

Downstairs, a different reality.

The data is stark. In the EU, 55% of large enterprises use AI. Among small businesses, the figure is 17%. That is not a technology gap. It is a class divide. And it maps almost perfectly onto how the technology is being understood and explained. Among small businesses that do use generative AI, only 29% are applying it to anything they would call a core activity. The rest are using it around the edges. Peripheral tasks. Domestic duties. The post.

Henry Ford wrote in 1922:

"Many people are busy trying to find better ways of doing things that should not have to be done at all. There is no progress in merely finding a better way to do a useless thing."

He was talking about building cars. He could have been writing this morning.

The distance between what is being built at the frontier and what is being adopted at the base is not closing. It is accelerating. And the way we talk about AI to mainstream audiences is making it worse. We are sending people downstairs and telling them it is the whole house.

There is a second problem. Possibly a larger one.

Most of what passes for "AI adoption" outside the LinkedIn bubble is being sold through the wrong door. The easy tasks. The obvious wins. Summarise this. Schedule that. The logic is that simplicity reduces friction. That people need a gentle on-ramp. That you show a small business owner how to draft a social post and then, gradually, they discover the rest.

It does not work like that. First impressions calcify. Show someone a powerful tool doing a trivial job and they will remember the trivial job. They will not come back looking for depth they were never shown existed.

I spoke recently with a small business owner who had tried AI, been walked through how to write copy and captions for their email marketing and quietly concluded it was not for them. That was months ago. They have not been back. That story is not unusual. It is the default.

The research confirms it. 82% of the smallest businesses cite a belief that AI simply is not applicable to what they do as the reason for not adopting it. Not cost. Not complexity. Irrelevance. A first impression, repeated at scale, until it became a conclusion.

The small businesses that could benefit most from AI are not reading the case studies. They are not at the conference. They are watching what their peers do. A baker. A joiner. A recruitment firm. A family-run logistics company. They take signals from people they recognise, doing things they understand, solving problems they actually have.

The onboarding narrative being built right now, the one that leads with email summaries and meeting notes and caption writing, is creating a first impression across millions of businesses that may never be revisited. That is not a slow adoption curve. That is a ceiling, being installed quietly, before most people have looked up.

Showing those peers doing genuinely surprising things with AI would move the dial far faster. And unlock value that the easy-wins playbook will never reach.

The irony is almost too neat. Within days of Ask.com's funeral, journalists were testing AI assistants to see if they would roleplay as Jeeves. They did. Willingly. Enthusiastically. "Very good, sir. I am entirely at your disposal."

Which tells you everything about where we are.

Upstairs, autonomous agents are rewriting how organisations think and operate. Downstairs, we are asking them to summarise the emails and write the captions. You can have Jeeves back. Better than ever. Faster, cheaper, tireless, available at 3am, never sulks.

Or you can ask the harder question.

What if this is not a better butler. What if it is a different kind of mind entirely and we have been wasting it on sorting the post.

Butler-ism (n.): The systematic reduction of artificial general intelligence to domestic task execution. Named in honour of Reginald Jeeves, who deserved better, and Ask Jeeves, which perhaps did not. Ask.com closed on May 1st, 2026. Jeeves' spirit, apparently, endures.

Asked at the counter

What is Butler-ism? The systematic reduction of extraordinary intelligence to domestic tasks. We have spent four years building the most capable reasoning systems in human history, and the dominant use case in many small businesses is "write me a summary of this email." It is the new Jeeves, wearing a different coat: summarise, draft, reformat, translate. It feels like efficiency. It is a misreading of what the moment requires.

Why do most small businesses fail to benefit from AI? Because first impressions calcify. AI is being sold through the wrong door, the easy tasks and the caption writing, so people see a powerful tool doing a trivial job and conclude it is not for them. 82% of the smallest businesses cite irrelevance, not cost or complexity, as their reason for not adopting it, and in the EU only 17% of small businesses use AI against 55% of large enterprises. That is not a technology gap. It is a class divide.

What is the difference between task-level and purpose-level AI use? Task-level use treats AI as a production accelerant: more output, faster, cheaper. In a world where everyone has the same accelerant, that is a commodity race won by whoever spends the most on compute. Purpose-level use puts AI upstream of execution, inside strategy, at the point where problems are still being defined. One gets replaced by a cheaper system. The other compounds.