AI-first is the discipline of flipping the default assumption about how work gets done. Instead of asking whether a human task can be automated, AI-first organisations require teams to prove a human is still required before adding headcount, budget or process complexity. It is a workflow redesign strategy, not a layoff strategy. And it is meaningfully distinct from AI-only — the substitution of AI for humans wholesale — which has now been shown, most publicly by Klarna, to produce lower-quality outcomes and force expensive reversals.
This blog unpacks how four companies — Duolingo, Shopify, Box and Klarna — are teaching the rest of us what AI-first actually looks like in practice, what the data says about productivity gains, and how to map your own owls and tutors before you make the most expensive mistake of 2026.
I speak only Swedish with my kids.
It's a small act of inheritance — the language of my parents and grandparents, carried into a Sydney household where the rest of life happens in English. I'm the human tutor in this story. I bring the mormor and farfar memories (technically farmor [father's mother]— my mother — but she always insisted mormor sounded cuter), the cultural texture, the occasional bit of discipline, and the lived warmth of a language passed down. But I'm not always available, not always patient, and frankly not as much fun on a Tuesday night as a green owl named Duo.
Duo is the supplement. He celebrates streaks. He gamifies vocabulary. He never rolls his eyes when Lucien mispronounces sjuksköterska for the eleventh time. (For the curious: it's roughly fhwook-SHUR-tair-ska, Swedish for "nurse" — and the infamous sj-sound shows up twice in a single word.) He turns adaptive, AI-driven repetition into something my kids actually want to do.
Together, Duo and I are extraordinary. Apart, neither of us would be enough.
A few months ago I put a question to a keynote audience in Singapore: what do tools like Duolingo do to human tutors? Augment them, or disrupt them? The question landed harder than I'd expected — because the same question is now being asked in every industry. Education. Financial services. Law. Marketing. Design. Software. The honest answer — the one most leaders aren't quite ready to hear — is both. And the professionals who'll thrive in the next decade aren't the ones who pick a side. They're the ones who learn to choreograph the dance between Duo and the human tutor.
There is a small yellow post-it on my desk that simply reads: Think AI First.
It feels almost paradoxical for a futurist to need a reminder. But the reminder is precisely the point. Our defaults are built for a world in which every task lived inside a human's head — and most of those defaults are now quietly out of date.
I dug into this recently with Travis Hess, CEO of Commerce, on his Keeping Commerce Weird podcast ahead of my AI keynote speech at Commerce Live in Chicago. One question kept surfacing through our conversation: what does flipping the default actually look like, in practice, inside a real organisation?
Last year, Duolingo's CEO Luis von Ahn put that idea to his entire company in a single rule. If a team wants more headcount, they must first present a case showing why an AI cannot do the work. The default has flipped. Instead of prove AI can do this before we automate it, the question is now prove a human is still required before we hire one.
That is the heart of what AI-first actually means. Not that humans are out. Not that AI replaces judgement. But that the starting assumption about how work gets done has changed.
Shopify's CEO Tobi Lütke made the same move with a memo he chose to publish rather than have leaked. "Reflexive AI usage" is now baseline at Shopify — the idea that reaching for AI should be the unconscious default, not a deliberate choice. Before any team requests more people or budget, they have to show why autonomous agents couldn't already be doing the work. Lütke describes AI as a multiplier on talent: his strongest performers, he says, are now shipping output their pre-AI selves would have considered impossible.
Aaron Levie at Box has the cleanest articulation of where most companies are going wrong. "Most companies are layering LLMs onto legacy workflows," he argues — bolting AI on top of processes that were designed in 1995. The point of being AI-first isn't to add a chatbot to your support page. It's to redesign the workflow as if AI were already in the room.
Developers using AI coding assistants completed tasks 55% faster than control groups in a study run jointly by GitHub and MIT, with pull request cycle times dropping from 9.6 days to 2.4 days at enterprise scale. Microsoft Research's randomised trial across 6,000 workers at 56 firms found employees saved roughly half an hour a week on email alone and finished documents 12% faster.
But the number that should sit on every executive's desk is from McKinsey's most recent global survey, published late last year. Nearly 90% of companies have now deployed AI in at least one business function. And 94% report no significant value yet.
That gap — between adoption and value — is what I have called The Framing Gap. It is not an indictment of the technology. It is an indictment of the strategy. Adoption is not transformation. Pilots are not redesign. Plugging ChatGPT into your existing process and hoping for compound advantage is the corporate equivalent of putting a Tesla electric motor in a horse-drawn carriage.
And while 94% of companies sit on that gap, the ones that don't are compounding away from them. Agentic AI capability is currently doubling roughly every three months — which means a one-year delay in AI adoption is no longer a 12-month gap but a 16x capability gap. The cost of waiting is no longer linear. It is exponential.
Here is where the story gets more honest than most AI-first pieces will admit.
Klarna, the Swedish fintech I've watched closely for years, became the global poster child for AI-first customer service. Their OpenAI-powered assistant was claimed to be doing the work of 700 full-time agents. Resolution times dropped by 82%. The company paused hiring for over a year and shrank headcount from around 5,500 to 3,400. By the numbers, it looked like a victory.
By the experience, it was a problem.
Last year, Klarna's CEO Sebastian Siemiatkowski admitted publicly what customers had already been feeling. Cost, he conceded, had become a "too predominant evaluation factor" in how the company organised support — and the result was lower quality service, falling customer satisfaction, and engineers being pulled in to answer tickets. Klarna is now rehiring human agents under a hybrid model. Gartner expects half of all companies that cut customer service jobs because of AI will need to rehire by 2027.
Klarna isn't a story about AI failing. It's a story about confusing AI-first with AI-only. They are not the same thing. Getting them confused is the most expensive mistake leaders are making in 2026.
Which brings me back to Duo and the Swedish kitchen table.
Duo is remarkable at the things Duo is good at: drilling vocabulary, gamifying repetition, lowering the activation energy to practise on a Tuesday night when Aurélien would rather be climbing on rocks. But Duo cannot tell Lucien why his mormor would have laughed at a particular phrase. Duo cannot read the room when Aurélien is tired and needs presence more than he needs another streak. That part is still my job. The warmth, the cultural texture, the human authority that turns a language drill into a relationship — those don't come from an app.
The question for every leader I work with is the same one I ask about my kids' Swedish.
Where in your organisation is the Duo — the work that should be reflexively, joyfully automated, freeing your people to spend their hours on what actually matters? And where is the human tutor — the work that requires judgement, empathy, presence, and the kind of authority only a person can hold?
Get that distinction wrong in either direction and you pay. Cling to the human tutor for everything and you will be outpaced by a competitor who lets AI handle the drills. Replace the human tutor entirely and you will end up like Klarna circa 2024 — efficient on a spreadsheet, hollow in the experience your customers actually remember.
The yellow post-it on my desk says Think AI First. The unspoken second line is the one most companies are still learning to write:
…not AI only.
The future of work isn't a contest between humans and machines. It's a choreography between them — so that we do less of the menial and the mundane, and more of the meaningful and the humane.
That, in the end, is the only AI-first strategy worth having.
AI-first means flipping the default assumption about how work gets done. Instead of asking whether a human task can be automated, AI-first organisations require teams to prove a human is still required before adding headcount, budget or process complexity. It is a workflow redesign strategy — most clearly articulated by Aaron Levie at Box and operationalised by Tobi Lütke at Shopify and Luis von Ahn at Duolingo.
AI-first is the discipline of designing workflows around AI from the start while preserving human judgement, presence and authority for the work that requires it. AI-only is the wholesale substitution of AI for humans, usually motivated by cost. Klarna's 2025 admission — that cost had become a "too predominant evaluation factor" in their AI-driven customer service strategy — is the clearest public example of why the two should not be confused. Klarna is now rehiring human agents under a hybrid model.
Yes. After publicly claiming its OpenAI-powered assistant was doing the work of 700 full-time agents and shrinking headcount from approximately 5,500 to 3,400, Klarna's CEO Sebastian Siemiatkowski conceded in 2025 that customer satisfaction and service quality had declined. The company resumed hiring human customer service agents under a flexible hybrid model. Gartner predicts that by 2027, half of all companies that cut customer service jobs because of AI will need to rehire.
The Lütke test, named for Shopify CEO Tobi Lütke's April 2025 memo, requires that before any team requests additional headcount or budget, they must first demonstrate why autonomous AI agents could not already be doing the work. It flips the historical default — prove AI can replace this human — to a new default: prove a human is still required. Lütke calls the underlying behaviour "reflexive AI usage" and has embedded it in Shopify's performance review process.
According to McKinsey's State of AI 2025 survey, nearly 90% of companies have deployed AI in at least one business function but 94% report no significant value yet. The cause is not the technology — it is the strategy. Most organisations are layering AI onto workflows designed for the pre-AI era, rather than redesigning the workflow itself. This pattern — what I call The Framing Gap — is the dominant reason AI investments stall.
As of 2026, agentic AI capability is doubling approximately every three months — a pattern I refer to as the New Moore's Law for agentic AI. The compounding effect means a one-year delay in AI adoption produces not a 12-month gap but a 16x capability gap relative to early movers. The cost of waiting is exponential, not linear.
Confusing AI-first with AI-only. AI-first treats AI as the new default for workflow design while preserving humans for work that requires judgement, presence and emotional intelligence. AI-only attempts to substitute AI for humans wholesale, typically to reduce cost. The latter strategy has now been shown — most publicly by Klarna — to produce lower-quality customer outcomes and force expensive reversals.
Anders Sörman-Nilsson is a Swedish-Australian futurist and the founder of Thinque, a strategic foresight and advisory firm. He is the author of three books — Digilogue, Seamless, and Aftershock — and serves as Adobe's AI brand ambassador and futurist-in-residence for the Brisbane Broncos and Barker College. His work on responsible and agentic AI has been featured by the Wall Street Journal, the New York Times, Forbes, Monocle and the BBC, and he has advised leadership teams at Apple, Google, Meta, McKinsey, LEGO, Dyson and Rugby New Zealand.
Subscribe to Decoding Tomorrow, his weekly newsletter on AI, ethics and the future of work, at anderssorman-nilsson.com.