Defining AI Ambition

On April 28, 2025, Luis von Ahn, chief executive of the language-learning company Duolingo, posted a company-wide memo on LinkedIn. Its headline sentence was that Duolingo “is going to be AI-first”. The memo was more specific than most such announcements. The company would gradually stop using contractors for work that AI could handle, a team would get new headcount only if it could not automate more of its work, and AI use would count in hiring and in performance reviews¹.

Outside the company, many readers heard only the first sentence, and they heard it as a plan to replace people. The backlash was loud enough that, within a few weeks, von Ahn posted again: “I do not see AI as replacing what our employees do,” adding that the company was still hiring at the same speed as before¹.

A year later, one of the memo’s concrete choices was quietly reversed. Employees had asked whether they were expected to use AI “for AI’s sake”. In April 2026 von Ahn told an interviewer that AI use was no longer a performance measure: what mattered was doing the job as well as possible, with AI where it helped².

The episode is not a verdict on Duolingo’s strategy, which only its leaders can judge. It shows two problems clearly. A two-word phrase carried very different meanings to different audiences. And one of the few measurable commitments in the memo counted how much people used AI, not what AI was supposed to change.

The core idea

An AI ambition is the choice of how much an organization expects AI to change its products, its processes, its workforce and its competitive position. It is a strategic decision, and it comes before anyone decides what to build.

The previous chapter, Start With Business Strategy, established where AI belongs: wherever it serves what the business must achieve. Ambition answers the next question, which is how far to go. A low-cost operator and a product innovator might both decide that AI matters to them. One may need AI to take cost out of processes it already runs; the other may need AI inside what it sells. Those are different ambitions with different bills attached.

The right ambition is not the largest one available. It is the one that fits the business strategy, the size of the opportunity, the organization’s readiness, the investment it can carry and the risk it is willing to accept.

A slogan becomes an ambition when it names choices

Compare two statements. The first: we will be the most AI-driven company in our industry. The second: over the next three years we will use AI to redesign our three highest-volume knowledge workflows and build it into two core products, and no AI system will make a final decision that affects a customer’s safety. The first wins applause. The second can be acted on the next morning.

The difference is that the second statement answers four questions. Where: which parts of the business should AI change, and which should stay mainly human-led? How deeply: a little cheaper, or fundamentally redesigned? How quickly: over what time horizon? How much: what investment and organizational change will the company accept? A good statement then adds a fifth element, a boundary: something the organization will not do, however capable the technology becomes. As What Is an Enterprise AI Strategy? showed, Richard Rumelt calls grand words without choices “fluff”³. An AI-first slogan with no answers to these questions is fluff of exactly that kind.

A useful way to hold the four questions is to think of a brief for an architect. Tell an architect only to “make it modern” and you will get handsome sketches, endless revisions and no idea of the cost. Tell the same architect to rewire the ground floor, refit the bathrooms, repaint upstairs, leave the roof alone, finish in six months and stay within a stated budget, and work can start. A real brief names the rooms, the depth of work in each, a date and a budget that matches the depth. Nobody moves walls on a repainting budget. And different rooms getting different depths is not a lack of ambition. It is a plan.

Adoption measures; ambition chooses

The second common confusion is between ambition and adoption. Adoption asks whether people are using AI: how many, how often, with which tools. Ambition asks how much the organization means to change because of it. The two move independently, and treating an adoption target as the ambition is a common mistake in AI planning.

The survey evidence shows how far the two have come apart. In McKinsey’s 2025 global survey, nearly two-thirds of organizations had not yet begun to scale AI across the enterprise, even though use in at least one function had become almost universal. The small group of high performers, as AI Is Changing Everything described, differed in intent: they were more than three times as likely as others to say they meant to use AI to bring about transformative change in their business⁴. The two figures count different things: the two-thirds is a share of all organizations surveyed, while “three times as likely” compares the high performers, about 6 percent of respondents, with everyone else. That is an association, not proof that ambition causes results. But it is consistent with a simple point: organizations rarely get transformation by accident.

High adoption with low ambition is easy to picture: almost everyone uses an assistant every week, and no process, role or product has changed. The opposite is just as real: a company that has committed to rebuild its core workflows, where usage is still low because the work has only begun. Duolingo’s reversal on performance reviews shows what happens when a usage measure stands in for the ambition. People optimize the measure. AI Strategy vs AI Adoption covers measurement in depth; an adoption target can support the ambition, but it cannot be the ambition.

Five levels of ambition, and higher is not better

Ambition comes in levels. They are best treated as a range of choices, not as a ladder that every organization must climb.

At the first level, efficiency, AI reduces the cost or effort of processes the organization already runs: drafting routine documents, summarizing, classifying. The process stays the same. At the second, augmentation, AI raises the quality or speed of people’s work through research help or decision support, and the person stays at the center. At the third, workflow transformation, the organization stops inserting AI into existing steps and redesigns the flow: AI assembles, drafts and recommends, and people handle judgment and exceptions. At the fourth, AI-enabled products, the question shifts from how the company operates to what it sells. At the fifth, AI-native, AI stops being a set of initiatives and becomes part of how the organization runs and competes. Marco Iansiti and Karim Lakhani cite Ant Group, built around data and algorithms from the start, as an example⁵.

Boston Consulting Group draws a similar line in its advice to executives. It distinguishes three plays: deploying AI for immediate productivity gains, reshaping critical functions, and inventing new products and revenue streams. Reshaping, it estimates, can raise efficiency and effectiveness in the targeted functions by 30 to 50 percent, against 10 to 15 percent for deploying tools⁶. Those are a consultancy’s planning ranges, not measured results from a published sample. In BCG’s 2025 survey of more than 1,800 executives, the leading companies put more than 80 percent of their AI investment into reshaping and inventing rather than into small productivity initiatives⁷.

None of this makes the top rung the right answer for everyone. A stable, tightly regulated business with limited AI opportunity may rightly choose efficiency and augmentation. A digital business in a market that AI is reshaping may need to aim much higher. As the previous chapter showed with Michael Porter’s generic strategies, a business that competes on cost will look first for efficiency, while one that competes on differentiation will care more about AI in the customer experience and the product⁸. And one enterprise rarely needs one level everywhere. Finance might aim for efficiency, engineering for augmentation, operations for workflow transformation and the product group for new offerings. That is usually more realistic than declaring a single level for the whole company.

Every level of ambition carries a bill

Each step up the range changes what the organization must invest, build and manage. The bill is not only money.

Efficiency needs tools, enablement and basic guardrails, and it can spread across the business with light central support. Augmentation needs AI built into the workflow, training and quality checks. Transformation needs process redesign, product management, data capability and AI engineering, which usually means dedicated teams rather than volunteers. AI-native needs reusable platforms, a strong data foundation, AI operations and often a redesigned organization. The AI Transformation Challenge introduced BCG’s rule of thumb that leaders put most of their AI resources into people and processes rather than algorithms⁹; the higher the ambition, the more that rule bites.

Exposure rises with ambition as well: dependence on suppliers, model errors, security, privacy, regulation, workforce disruption and reputation. So ambition, risk appetite and the capacity to govern must move together. Two companies can see the same opportunity and decide differently. One accepts AI recommendations that a person approves; the other insists on fixed rules for that process. Neither is automatically wrong. How to tier and control those risks is the subject of Module 07.

The practical test for any strategy review is a single question: does our investment match our stated ambition? A strategy that claims transformation while funding scattered pilots has a gap that can be read straight from the budget. If the words say transformation and the budget says pilots, the organization does not yet have an ambition. It has a slogan.

Match ambition to readiness, then close the gap

Ambition also has to meet the organization as it is. Readiness here means the state of data, technology, talent, governance, leadership and the capacity to absorb change. Placing ambition against readiness gives four positions, and each calls for a different move.

With high ambition and high readiness, the task is to move faster and scale what works. With high ambition and low readiness, the task is to build capability. This position is easy to misread. Low readiness is not a reason to lower the ambition; it is a reason to build what the ambition needs, in stages, and to fund each stage properly. With low ambition and low readiness, the first job is to clarify the strategy: decide what AI is for before funding it. And with low ambition and high readiness, the job is to test the caution. If AI could lower barriers to entry in the market, let a rival operate at much lower cost or change what customers expect, a modest ambition is itself a strategic risk.

The distance between the ambition and today’s capability is the ambition gap. A large gap is allowed. A strategy that does not explain how the gap will close is not. Ambition can also follow a trajectory: efficiency and augmentation now, transformation next, new products later. That echoes the three horizons of growth described in What Is an Enterprise AI Strategy?¹⁰. How to assess readiness in detail is the subject of The AI Maturity Model.

Large gaps have been closed before. In 1989 Gary Hamel and C. K. Prahalad described how the equipment maker Komatsu set out to “encircle Caterpillar” when it was less than 35 percent of its size, then pursued that single ambition through a succession of funded medium-term programs, each with its own target¹¹. That is what building capability in stages means: keep the ambition, and attach money and a target to each stage.

Story: Sanofi goes “all in”, area by area

In June 2023 the French drug maker Sanofi announced that it was going “all in” on artificial intelligence and data science. Its chief executive, Paul Hudson, said the ambition was “to become the first pharma company powered by artificial intelligence at scale”¹². On its own, that sentence is the kind of phrase this chapter warns about. What makes the case useful is the record of choices behind it, in the company’s announcements and in its own and its partner’s annual filings. Those choices did not ask the same thing of AI everywhere.

For every employee, the move was broad and shallow. In 2023 Sanofi rolled out plai, an AI-driven app built with the AI company Aily Labs that brings company data into one view, to more than 23,000 internal users¹³. That is augmentation: cheap per person, quick to deploy, modest in depth. In manufacturing and supply, the aim was efficiency in processes the company already ran. Sanofi built its own AI-enabled yield optimization, which learns from past batches to raise yields and save raw materials, and it reported that the app’s supply-chain use could predict 80 percent of low-inventory positions¹³¹². These are the company’s own capability claims; it has not published audited savings from them.

Research carried the largest ambition, and there the money was staged. Instead of building AI discovery alone, Sanofi signed partnerships with specialists. The largest, with Exscientia in January 2022, covered up to 15 drug programs in oncology and immunology and carried a headline of up to 5.2 billion dollars in milestone payments. The money committed at signing was 100 million¹⁴. By the end of 2025, after Exscientia had become part of Recursion, the partner reported 4 million more for expanding the deal and roughly 37 million in milestone payments, with the programs still working toward their first development-candidate milestones¹⁵. About 140 million had been paid against a 5.2 billion headline, under 3 percent. The rest is paid only if the science works.

The boundary came early. In April 2023 Sanofi set up a cross-functional working committee on responsible AI, and it states that “a human must be involved and is always accountable for the outcomes of an AI system”¹⁶.

The record does not show whether these bets will pay off. Drug discovery takes a decade, and the evidence on AI is still mixed: in the first analysis of AI-discovered molecules in clinical trials, they succeeded in Phase I 80 to 90 percent of the time, well above historical averages, but in Phase II at about 40 percent, roughly the industry norm¹⁷. The case is not proof that Sanofi chose well. It shows what an ambition looks like once it is written as choices: a different depth in each area, money that matches each depth, the biggest bet paid for in stages, and one line the company says it will not cross.

What this means for leaders

An AI ambition is a short document with large consequences. It tells the organization where AI should change the work, how deeply, by when and at what cost, and it tells everyone what will not happen. Writing it is the executive team’s job, not the technology function’s, because it commits the business, not the IT budget.

Three habits make it hold. First, set the ambition area by area rather than as one level for the whole company, and say so openly. Second, read the budget as a test of the words: if the money cannot carry the stated depth, either fund it or state a smaller, true ambition. Third, when readiness is low, keep the ambition and stage the capability, as Komatsu did and as Sanofi’s milestone-funded research deals do, instead of pretending to be ready or quietly lowering the sights.

Check yourself

1. A higher AI ambition means a better strategy.MythFact
2. Setting a target for AI usage is a way of setting the AI ambition.MythFact
3. Different parts of one company can hold different AI ambitions.MythFact
4. Low readiness means the organization should set a low ambition.MythFact
5. Leading companies in BCG’s 2025 survey put most of their AI investment into reshaping functions and inventing offerings.MythFact
6. AI-discovered drug candidates have already proven more successful than traditional ones across all trial phases.MythFact

Reflection: read your own ambition

What comes next

An ambition tells the organization how far to go. It does not yet say which opportunities deserve the money. The next chapter, Finding Strategic AI Opportunities, shows how to identify the opportunities worth executive attention, instead of relying on random lists of use cases or the latest technology trend.