What Does ‘AI-First’ Mean? Duolingo Declared It in April. By June It Was Explaining Itself.

What does AI-first mean? An AI-first company makes AI the default way work gets done, and asks people to justify the exceptions. Duolingo’s April 2025 memo spelled out what that meant there: gradually stop using contractors for work AI can handle, look for AI use in hiring, count it in performance reviews, and add headcount only where a team can’t automate more of its work. Users and staff heard replacement. Within weeks the chief executive was explaining what the memo had not meant.

Duolingo did not publish a savings number, a layoff count or a measured result. It published a sentence — we are going AI-first — and spent the following weeks explaining what the sentence had not meant. That makes it a weak case study as evidence, and one of the most useful as a warning.

The memo: April 28, 2025

On 28 April 2025, Duolingo's co-founder and chief executive, Luis von Ahn, told staff the company would become AI-first, in an all-hands email the company posted publicly. The constraints were specific: gradually stop using contractors for work AI can handle; look for AI use in hiring; count AI use in performance reviews; grant headcount only where a team cannot automate more of its work; and have most functions run initiatives to change how they work fundamentally. The memo argued that AI would let the company create teaching content at a scale manual work could not reach, and insisted it was about removing bottlenecks so employees could focus on creative work and real problems, not about replacing its people.

The public read the constraints, not the reassurance. Coverage in The Verge and elsewhere led with contractors replaced by AI. Users threatened to cancel subscriptions, in comments the reporting quoted directly: people heard replacement, whatever the memo meant to say. For a consumer company whose product asks learners to trust it daily, that hearing is not a communications footnote. Attention is the budget here too — the firm spent weeks of it explaining a sentence instead of teaching anyone anything.

The clarification: May and June 2025

Within weeks von Ahn was clarifying. In a follow-up reported by Fortune in late May, he wrote that he does not see AI as replacing what employees do, described it as a tool to accelerate work at the same or better quality, and said the company was continuing to hire at the same speed as before. By June, in an interview reported by Fortune on 9 June, he said he had not expected the amount of blowback, that he should have been clearer, and that every tech company was doing similar things but Duolingo had been open about it. Later in the summer he said the memo had lacked context and that no full-time staff had been laid off because of it.

So the honest inventory is this: an announcement, a backlash, and a retreat into explanation. No hard outcome number exists — no published saving, no productivity measure, no count of contractor work actually transferred, nothing a reader could verify or copy. Any headline that attaches one is inventing it. What the episode does measure, roughly, is the cost of language. AI-first, as a phrase, told staff to wonder about their jobs and customers to wonder about the product, before a single workflow had been redesigned.

A usable definition of AI-first

As Duolingo wrote it, AI-first names a priority, not a use. It tells people AI comes first without saying what it is for, what it will never do, or what stays with people. A usable version, in plain operating language, would have sounded like this: AI will draft the routine exercises; teachers will set the standard, judge the exceptions and own the learner relationship; no role changes until a pilot proves the finished work is as good, timed and checked. Same direction, completely different hearing, because people can see where their attention goes and where the machine stops.

That order, workflow first and phrase last, is the mirror image of IBM’s HR chatbot rollout. IBM made the bot the front door and paid for it with a minus-35 score until training caught up. Duolingo made the phrase the front door and paid for it in public trust before any tool was judged. In both, the announcement ran ahead of the operating habit. The readiness question for a small firm is therefore not whether you are AI-first. It is whether you can name, this week, one routine workflow the machine will absorb, one judgement your people will keep, and the sentence you will say to the team that matches that reality. Our AI readiness assessment starts exactly there, long before any slogan, and the MIT article covers what happens to firms that announce transformation before the pilot earns it.

Other AI-first companies: Shopify, Krafton and Klarna

Duolingo wasn’t alone. Three other public examples show the same idea written different ways:

• Shopify (CEO memo, posted publicly by the CEO and reported by CNBC and The Verge, April 2025). “Reflexive AI usage” became a baseline expectation for everyone. AI use was added to performance and peer reviews, and teams had to show why they couldn’t get work done with AI before asking for more headcount. It made the same moves as Duolingo, but it was aimed at how staff work, with no line about replacing an outside group.
• Krafton, the maker of PUBG (company announcement, reported by The Verge, 23 October 2025). It declared an “AI First” strategy built on agentic AI to automate work, a dedicated GPU cluster, a yearly budget to put AI tools in employees’ hands, and HR policies rewritten around AI. The message to staff was investment in their tools, paired with a promise that people would focus on creative and complex work.
• Klarna (company spokesperson, reported by Fortune). After hiring people back into customer service because a focus on cost had hurt quality, Klarna said it was “very much still AI-first”, with the focus moving from cutting costs to improving service. Read the Klarna rehiring story.

All of them made AI the default. What differed was whether the people affected could see what stayed theirs.

How a small firm can go AI-first without the backlash

1. Map one workflow before you name a strategy. Pick a routine job, such as order entry, invoicing or scheduling, and write down every step. A process map shows what AI can take and what needs a person.
2. Write down what stays with people. Judgement calls, client relationships and final sign-off. Say it before anyone asks.
3. Pilot quietly and measure finished work. Count review time and rework, not just drafting speed.
4. Tell the people whose work changes first, in person. That includes contractors and part-time help, not just staff.
5. Put it in writing once it works. A short AI policy beats a slogan, and our AI rollout guide covers the announcement itself.

The aim isn’t a firm where people operate AI tools all day. It’s one where the routine information flows through a loop the machine runs, and people spend their time on the work only they can do.

What to do instead, in one paragraph

Run the pilot quietly. Train the three people it touches. Measure finished work, not enthusiasm. Then tell the team what actually changed, with the numbers you actually have, and what will not change. If the result is good, the phrase you need is not AI-first. It is the name of the task, the hours returned, and where those hours now go. People can support a sentence like that because it respects the scarce thing they were afraid of losing: their attention, and the standing it represents.

The contractor line deserves its own sentence

Notice which constraint travelled furthest. It was not the performance-review line or the hiring line, both internal. It was the promise to stop using contractors for work AI can handle — a group with no seat at the all-hands, reading about their future in a public post. In a small firm the equivalent audience is not a platform's user base. It is the bookkeeper who comes in two days a week, the seasonal help you rely on every busy month, the freelancer your clients know by name. If a sentence about automation reaches them before a conversation does, you will spend your own June explaining April. Tell the people whose work changes first, in person, with the pilot result in hand — and tell them what stays theirs.

Planning to tell your team you're going AI-first? Run the wording past Ada first. She will test it against your real workflow and build a calmer rollout into your free report at /audit/. Duolingo’s lesson is free if you take it before your announcement, and expensive only if you take it after. Want help planning the rollout itself? See how we work with small firms.

Frequently asked questions

What does AI-first mean?

An AI-first company makes AI the default way work gets done and asks people to justify the exceptions. In Duolingo's 2025 memo that meant phasing out contractors for work AI can handle, looking for AI use in hiring, counting it in performance reviews, and adding headcount only where a team cannot automate more of its work.

What did Duolingo's AI-first memo say?

On 28 April 2025 chief executive Luis von Ahn told staff Duolingo would become AI-first. The memo said the company would gradually stop using contractors for work AI can handle, look for AI use in hiring, include AI use in performance reviews, grant headcount only where a team cannot automate more, and have most functions change how they work.

Did Duolingo lay people off because of AI-first?

Von Ahn later said no full-time staff had been laid off because of the memo, and in May 2025 he said the company was continuing to hire at the same speed. The memo did say Duolingo would gradually stop using contractors for work AI can handle. The company published no count of contractor work actually moved to AI.

Should a small business call itself AI-first?

The label matters less than the plan behind it. Before you use the phrase, name one workflow AI will take on, the judgement your people keep, and what will not change. If you can say that in one sentence, you may not need the slogan at all.

How do you announce AI to your team without a backlash?

Run a pilot first and measure the finished work. Tell the people whose work changes, including contractors, in person and before any public post. Say what AI will do, what stays theirs, and what the pilot actually showed, using only the numbers you have.

Sources: The Verge, Duolingo will replace contract workers with AI, 28 April 2025; Fortune, Duolingo CEO walks back AI-first comments, 24 May 2025, and Duolingo’s CEO surprised by backlash, 9 June 2025. Shopify: CNBC, 7 April 2025 and The Verge. Krafton: The Verge, 23 October 2025. Klarna: Fortune, as cited in our Klarna article. No hard outcome figure was published by Duolingo, and none is claimed here.