Remember the metaverse? It was supposed to change everything. Spoiler: it didn’t.  AI, on the other hand, quietly is.
 
COVID-19 is old news when discussing office demand.  Now, everyone is wondering how AI will affect office usage (i.e., headcount).  Will AI simply replace large quantities of employees? Or will AI augment productivity while keeping headcount the same? 

It’s too soon to tell.  In the meantime, here are 3 takeaways from The Economist article below on how AI is already reshaping the office:

— Everyone’s Using It (Even If Quietly): From meeting summaries to writing help, AI is in the workflow. Some folks experiment openly, others keep it under the radar.

— Tasks Are Being Delegated: We’ve moved from “help me think” to “just do it for me.” AI’s becoming the default assistant.  This shift will

— Office Vibes Are Shifting:  Meetings aren’t just conversations anymore. They are part of a historical account.  AI is tracking to-dos and accountability.  Performance reviews might soon track AI usage.

No, it’s not a revolution yet—but AI is part of the office from here on out. 

The next big thing doesn’t always turn out that way. There was a spasmodic moment in the early 2020s when the metaverse was going to be the future. A McKinsey report published in 2022 reckoned that it could generate up to $5trn of value by 2030. Meta got its new name. Some people were appointed chief metaverse officers. A few of them may still be out there, banging on about non-fungible tokens and how Barbados has an embassy in Decentraland. Everyone else has moved on.

Generative AI is plainly not the metaverse. It may end up falling short of the headiest expectations. Its full impact will become clear only over a long period of time. Many firms say they are disillusioned with their returns to date. But the office is already a different place because of the technology.

The evidence for that is partly quantitative. Employees are often adopting the technology unilaterally, working out for themselves how best to use it. Some are doing so surreptitiously, uncertain whether they will get credit or replaced. But the firms behind the frontier models can see what’s happening.

In a new NBER study, a team of researchers from OpenAI, with David Deming of Harvard University, document how people use ChatGPT. Although personal use of OpenAI’s chatbot has grown even faster, the daily average number of work messages zoomed from 213m in June 2024 to 716m a year later.

The latest version of the Anthropic Economic Index, a piece of analysis by researchers at the firm behind Claude, distinguishes between “automation” and “augmentation” modes—one being a more directive interaction in which a user tells the model to do something, the other a more collaborative pattern of questions and feedback. For the first time in the index’s short life, instances of automation outstripped those of augmentation, suggesting that ever more tasks are being delegated to AI.

Your own eyes and ears provide more proof that generative AI is increasingly part of office life. AI provides a constant background hum to work conversation. If you overheard someone asking “How do you use it?”, you’d know what was being referred to. Meetings now routinely end with phrases like “I guess I still have a job, then,” or “It’s really the next generation I worry about.”

The jargon is inescapable. People who have no clue what they are talking about are throwing around words like “alignment”, “non-determinism” and “agentic”. The in-crowd always have new ways to signal their credentials. First it was RAG, now it’s MCP. Did you listen to the new Dwarkesh, by the way?

The assumption that AI is everywhere is slowly taking hold. Meetings are routinely transcribed and summarised by a machine. You are no longer having a discussion with colleagues; you are part of the historical record. Usage of AI may well be part of how your performance is assessed. Some firms have dashboards to monitor employees’ adoption of the technology. Most bosses will have sent out the message that they expect staff to experiment with it.

Some basic premises are breaking down. That interviewee on your screen, for example. Jonathan Black, the head of the careers service at Oxford University, recounts the story of a job candidate asking an employer to repeat the question because the computer didn’t hear it. To catch more accomplished cheaters, AI invigilation services monitor if someone is switching tabs before they give an answer, or keep track of the time taken to answer a question.

Writing-related requests are the most common use of ChatGPT at work, according to the new study. That may well mean you are encountering fewer grammatical errors and more factual ones. You are also more likely to be reading, or indeed producing, generic content. Sterile language has long been part of workplaces: think airline-safety demos or call-centre messages insisting that “your business matters to us” (even if your time patently doesn’t). All correspondence is now lightly chlorinated.

None of this yet amounts to transformation. But AI is leaving its mark on workplace behaviour, language and assumptions. The metaverse it ain’t. ■

Subscribers to The Economist can sign up to our Opinion newsletter, which brings together the best of our leaders, columns, guest essays and reader correspondence.