Every business in the world today wants to be AI-first. Everyone wants to maximise productivity and get machines to do repeatable tasks. This lovely essay by Aaron Horwath begins with one such firm.
Before Lukas Weber writes a word, he checks in with the small army of AI agents he has built: one has combed the week’s industry news and customer calls for newsletter topics, another has drafted his replies to unread emails, another has flagged concerning trends in a campaign. “The expectation from our executives is I shouldn’t write at all,” Weber says. “I manage the agents and check their work before anything goes out.” Multiply Weber across his firm, then across every competitor, and you start getting the idea of the point Horwath is trying to make.
Horwath, who runs AI operations at a creative technology company, describes what every AI-bullish executive calls success: “everyone using AI for everything; human knowledge workers sitting above teams of agents all built on the same small ecosystem of LLMs.” The catch is that the competitors are doing exactly the same. “It’s a funny predicament: all day, humans across the same industry – ostensibly hired for their unique critical thinking and experience – collectively pump every thought, business challenge, idea and solution through the same set of AI tools. Almost as if all these companies have simultaneously hired the same single employee: Mr. Claude (or one of his contemporaries).” The consequence: “when every company in an industry leverages essentially the same single brain indiscriminately, competitive differentiation erodes. The value proposition of companies paid for solutions only they can provide collapses.”
The first symptom is a small one. “The rise of Claude and ChatGPT inside knowledge work has resulted in an unlikely victim: the em dash. Beloved by writers, LLMs use them ad nauseam, and they have quickly become markers of ‘AI slop.'” But the punctuation is a clue to something larger, which Horwath calls “airbrain,” after Kyle Chayka’s “Airspace” – the way Instagram-era algorithms once flattened the world’s cafes into one look, “Minimalist furniture. Craft beer and avocado toast. Reclaimed wood,” until “changing places can be as painless as reloading a website.” Now, he writes, “Newsletters, websites and ads across competitors and industries are converging on the same Claude-aesthetic, the same tone of voice and the same way of articulating ideas,” and “LLMs have processed them out to create a generic bleh, with one company’s output increasingly indistinguishable from another.”
This is no accident, since “LLMs are prediction machines at the end of the day, taking a statistical guess at the likeliest next word, or token.” Horwath cites a 2026 study by Duke’s Emily Wenger and the Technion’s Yoed Kenett that tested 22 commercial models against more than 100 people: individually the models sometimes won, but “as a population, the LLMs converged on strikingly similar answers, while the humans didn’t.” Overreliance, Wenger concluded, “will smooth the world’s work toward the same underlying set of words or grammar, tending to make writing all look the same.”
“But not all efficiency is created equal, a lesson the Army Corps of Engineers, back in 1962 in central Florida, learned the hard way.” To stop the Kissimmee River flooding, they straightened its 103 miles of bends and dredged it deep; the flooding stopped, but so did the water’s oxygen, and some 35,000 acres of habitat were lost before a $1 billion project began, decades later, to put the bends back. Work inside a company has its own bends, and AI promises to straighten them: “the result is a transformation of knowledge work into something more akin to manufacturing, with humans operating as the foremen keeping an eye on the metaphorical assembly line. But manufacturing is no place for innovation. No one wants the person on the Ford assembly line coming up with new ways of screwing in the driver’s seat.” Those bends “served a critical purpose. Manually developing ideas helped refine and nurture innovation through varying perspectives.”
For anyone who spends their days hunting durable competitive advantages, the next passage is the essential one. “Ultimately, an organization’s Unique Selling Points are the aggregate unique talents of all the people they hire. Building an organization with the right combination of people is, and always has been, the ultimate moat. To then shove this all through AI is to filter out all the talent you’ve paid for.” Horwath enlists strategy professor Jay Barney, writing in MIT Sloan Management Review: “How can AI be the centerpiece of a sustained competitive advantage when everyone has it? … The value that AI unlocks will be unlocked for all. … Far from being a source of differentiation, artificial intelligence will be a source of homogenization.”
There is a cost to the worker too. “A 2026 Nature report found it takes shockingly little time for skills to atrophy once AI has been introduced,” and in a survey of 2,500 employees, “39% said their reliance on AI was actively making them less intelligent.”The bill for the firm arrives later: when companies “eventually wish to return to human-generated solutions and content, they might find they need time to rebuild the skills that most directly contributed to their differentiation in the first place.”
Horwath’s wager is that sameness stops paying, as it did for the cafes. Once they had become interchangeable, “to stand out in an increasingly competitive market, cafés are turning their attention to offering differentiated experiences,” because “the novelty of a product is now just as important as the product itself.” Knowledge work will follow: “once every player in an industry is leveraging AI, any competitive advantage it offers is neutralized; they will return to uniqueness – in content, in strategy, in solutions.” Even the big consultancies agree; Deloitte’s 2026 report holds that “Technology – especially something as increasingly ubiquitous as AI – is replicable. People aren’t”.
The shape of that future, he thinks, is a barbell. Drawing on a 2026 Cambridge paper by Erin McGurk and David Khachaturov, “on the far left of the barbell will be AI-produced assets with near-zero marginal cost. At the other far end will be products of verifiably human origin that sell at a premium. Most importantly, the middle will hollow out.” The oversight jobs drift to cheaper markets, since “what sense does it make to pay a premium for oversight when a worker with clear instructions and a Claude account can produce or evaluate that same, homogenized output for a fraction of the cost?” A smaller elite of tastemakers and strategists will be judged on “quality and novelty,” not output volume. Tellingly, the labs are already hiring them: “Anthropic is hiring highly paid tastemakers, inoculating itself against the very airbrain it’s creating.”
“The future of knowledge work, then, will be ruthlessly bifurcated.” And even that may not be the end, once some executive “buzzing on Zyn and pounding cans of Liquid Death” notices that his surviving tastemakers need not be full-time and turns to gig staffing (Upwork reports “skilled freelancing rose from 28% of skilled knowledge workers in 2025 to 38% in 2026”). That, Horwath argues, mistakes a task for a job. He borrows Rory Sutherland’s “doorman fallacy”: “opening the door is only the notional role of a doorman; his other, less definable sources of value lie in a multiplicity of other functions, in addition to door-opening: taxi-hailing, security, vagrant discouragement, customer recognition, as well as in signaling the status of the hotel.” The verdict is blunt: an organisation “composed of overwhelmingly fractional workers isn’t an organization at all. (And perhaps neither is an organization that has outsourced too much of its thinking to LLMs).”
The essay ends not with a forecast but with Weber. “There are tasks AI handles now that I never wanted to do,” he says, “But I do miss the actual act of writing and having the product be uniquely mine.” The future Horwath sketches gives that feeling back its value, though to a smaller group, in “a return to executing work that is valued because it is uniquely his,” and human.
Disclaimer : “Meta Platforms Inc. (owner of Instagram) forms part of the active holdings within the Marcellus Global Compounders Portfolio, a strategy offered by the IFSC branch of Marcellus Investment Managers Private Limited and regulated by the IFSCA. Accordingly, Marcellus, its employees, their immediate relatives, and clients may maintain interests or positions in these securities. Any reference to these companies or products is intended strictly for informational and educational purposes within the context of this discussion and should not be construed as investment advice.”
If you want to read our other published material, please visit https://marcellus.in/blog/
Note: The above material is neither investment research, nor financial advice. Marcellus does not seek payment for or business from this publication in any shape or form. The information provided is intended for educational purposes only. Marcellus Investment Managers is regulated by the Securities and Exchange Board of India (SEBI) and is also an FME (Non-Retail) with the International Financial Services Centres Authority (IFSCA) as a provider of Portfolio Management Services. Additionally, Marcellus is also registered with US Securities and Exchange Commission (“US SEC”) as an Investment Advisor.
