AI Is Not a Strategy

There is a strange tension in corporate America’s relationship with artificial intelligence right now. Companies are under enormous pressure to demonstrate that they understand how consequential the technology is, while there is considerably less agreement about what, exactly, they should be doing with it. The result has been a familiar feature of technology cycles: the appearance of innovation sometimes moves faster than the underlying innovation itself.

In a recent New York Times opinion piece, Julie Averill, the former CIO of Lululemon, takes aim at precisely this phenomenon. Its critique is not that AI is useless or that companies should ignore it (she believes the opposite, as do we), but that AI has become sufficiently important to investors, boards and the broader market that there is now an incentive to present most ordinary corporate decisions through an AI lens. Existing automation gets relabeled as AI. Cost reductions become AI efficiencies. Companies announce themselves as AI-first before the underlying operations have materially changed. Perhaps the most revealing example is layoffs: some companies have reduced headcount in anticipation of productivity gains from AI systems that have not yet been fully developed or deployed. The organization, in other words, is being redesigned around an efficiency that management expects the technology to eventually produce.

The phenomenon has acquired the name “AI-washing.” It is an extension of a problem that predates generative AI, in which companies learn the vocabulary of a technological shift faster than they learn how to extract value from it. The incentives are not difficult to understand. Telling investors that a company is becoming leaner because management overhired, missed its targets or simply wants to improve margins is a very different story from saying that artificial intelligence has made a new operating model possible. The latter sounds like innovation. It also suggests that the company is on the right side of a technological transition rather than cleaning up decisions made during the last one.

A Wall Street Journal article recently offered what looks at first like a very different story, but is really an argument for much of the same discipline: an announcement that Chili’s (yes, the restaurant chain) has deliberately declined to go “all in” on AI. Its technology team brainstormed dozens of possible applications and found only a handful worth seriously investigating. Meanwhile, CIO Chris Caldwell has spent much of the past two years investing in decidedly unglamorous technology: replacing outdated devices, rebuilding Wi-Fi across roughly 1,200 restaurants, improving payment and kitchen systems, and fixing technology that employees and customers actually encounter every day. Chili’s has experimented with AI where the use case makes sense, including inventory forecasting and replenishment, but Caldwell’s standard is straightforward: does the technology materially improve the business or the customer experience? If it does not, the fact that it contains AI is beside the point.

These articles are not really making opposing arguments; they are describing two sides of the same problem. Averill  is concerned with companies using AI as a narrative before it has become an operating reality. Caldwell offers an example of what happens when a company refuses to do that. Both point toward a principle that should be obvious but has become surprisingly easy to lose in the current environment: technology strategy should begin with the needs and constraints of the business, not with the technology that happens to be receiving the most attention.

That principle is worth defending. But it also raises a question that gets considerably less attention: what does it mean for a company that already exists, and what does it mean for one that is being designed today?

AI has amplified an old corporate innovation problem

Long before generative AI, corporate innovation had a tendency to begin with the mechanism rather than the problem. A company decided it needed an innovation lab, an accelerator, a venture program, a digital transformation initiative or some other visible expression of innovation, and only afterward confronted the harder question of what meaningful business problem the machinery was supposed to solve.

AI has not created that tendency, but it has certainly accelerated and intensified it.

When leadership decides that the organization needs an “AI strategy,” the instruction itself can subtly determine the answer. Teams go looking for AI use cases because AI use cases are what they have been asked to find. Boards request AI roadmaps, executives announce AI initiatives and business units inventory tasks that might be automated. The technology becomes the premise, and the business problem has to be found afterward. The resulting pilots may work perfectly well and still be largely irrelevant to the company's most important constraints or opportunities. An organization can become quite sophisticated at deploying AI without becoming meaningfully more innovative.

Our work at Misfit Labs has spanned both sides of this problem. We work with established companies looking for new sources of growth and innovation, and we build companies from the ground up. The contexts are different, but the useful starting point is remarkably similar: begin with the problem, understand why it remains unsolved, and only then determine what combination of technology, people, capital and organizational design gives you an advantage in solving it.

Sometimes AI fundamentally changes that answer; sometimes it does not.

Chili’s example is instructive precisely because there is nothing especially sophisticated about the logic it applied. If restaurant connectivity is unreliable, employees are working on outdated devices, or ordering and payment systems create friction for customers, those are technology problems worth solving. An AI application that does not address one of the company’s meaningful constraints may be technologically impressive while remaining strategically irrelevant. Conversely, replacing Wi-Fi access points may be utterly boring and enormously valuable.

But starting with the problem does not necessarily mean thinking incrementally about the solution. This is where we think some of the discussion about corporate AI stops too early. The question is not simply whether AI can make an existing process cheaper or faster but rather whether a meaningful change in technological capability can alter what a company is capable of doing in the first place.

We explored one version of this in an earlier blog post, The $1.4 Billion Lesson, in which IKEA treated automation as the beginning of an organizational redirection, rather than its conclusion. It used some of the capacity created by automation to move employees into a different customer-facing business, launch a new business unit, and generate new revenue. The technology changed the economics of existing work; management decided what to build around that change.

This is where the distinction between efficiency and innovation becomes important. If AI reduces the cost of an existing workflow by 20 percent, there may be a perfectly worthwhile efficiency opportunity. But if the same technological change makes it economical to serve customers who were previously too expensive to serve, allows expertise trapped inside the organization to be deployed in a new way, makes a previously impossible product feasible, or creates an entirely new source of revenue, the opportunity is no longer adequately described as efficiency.

This is also why layoffs made in anticipation of future AI efficiencies are so revealing. There will certainly be legitimate cases in which a company redesigns work around AI and concludes that it requires fewer people. But the sequence matters. There is a meaningful difference between understanding how the work has changed and building an organization around that reality, and reducing the organization first because management assumes that some future technological capability will eventually fill the gap. One is organizational redesign. The other is an organizational hypothesis presented as though the redesign has already happened.

The broader point is that technology creates possibilities. It does not make the organizational decision for you. When AI changes the economics of work, leadership still has to decide whether to take the savings, increase output, redeploy expertise, build a new capability, enter a new market or do some combination of these things.

In our corporate innovation work, this is often where the more interesting opportunities begin to emerge. Sometimes the right answer is to improve an existing process. Sometimes it is to do an audit and correct workflows. Sometimes it is to build entirely new capability inside the organization. And sometimes the technological change makes something possible that sits sufficiently far outside the existing operating model that it deserves to become a new product, business line or company altogether. The discipline is the same in each case: technology should expand the set of possible answers, it should not predetermine the question.

The advantage of having nothing to transform

For an existing company, all of this happens inside an inherited system. There are employees, customers, processes, technology stacks, management structures, incentive systems, contracts, physical assets and accumulated institutional knowledge. Some of those things are inefficient. Some exist because of technological limitations that no longer apply. Others may appear inefficient while performing functions that are difficult to see from a spreadsheet. Changing one part of the system can create consequences somewhere else.

For these companies, the discipline advocated by both the aforementioned articles seems headed in the right direction: start with the business, understand the actual constraint, determine whether AI, another technology, a process change or no technological intervention at all is the best answer. The goal is not to become maximally AI-enabled; the goal remains to build a better company.

A company being formed today gets to apply exactly the same principle under radically different conditions. It does not have to decide how AI fits into an organization built before AI existed. It gets to decide what organization should exist given that AI does.

That difference is more important than it sounds at face value. A founder does not need to ask how to automate a workflow inherited from 2005 if there is no reason to create that workflow in the first place. She does not need to make a traditional department 30 percent more productive if the capability that department provides can now be assembled differently. She does not have to decide how many existing jobs AI might eliminate because she has not created those jobs yet.

This is the distinction between optimization and architecture: the incumbent is generally deciding which parts of an existing system should change, but the founder has the advantage of deciding what the system should be.

This is also where the conversation about AI-native companies can become confused. Being AI-native should not mean using AI everywhere. If anything, a startup that begins every decision by asking “How can we use AI here?” is committing the same error as the incumbent engaged in AI-washing. It is allowing the technology to dictate the problem.

An AI-native company should instead be native to the technological reality in which it is being built. That means understanding which capabilities have become dramatically cheaper, which forms of expertise remain scarce, where human judgment becomes more important rather than less, which functions actually need to exist inside the company, and how much organizational machinery is required to turn those capabilities into something customers want.

Sometimes the answer will involve extensive use of AI. Sometimes the right technology will be something else entirely. What matters is that the company is not unconsciously reproducing an organizational architecture designed around constraints that no longer exist.

Efficiency is a smaller idea than redesign

This distinction also changes how we think about the economics of AI. Much of the current corporate conversation is understandably focused on productivity: can the same company produce the same output with fewer people or lower costs? Those are important questions, particularly for established businesses with large existing cost structures. But they are fundamentally questions about making an existing model more efficient.

In an earlier blog post, we argued that the more consequential possibility is that the model itself changes. In What Capital Efficient Growth Actually Looks Like Right Now, we looked at the relationship between capital, headcount and growth. For much of the venture era, raising money was how startups purchased organizational capability. A company raised a round and hired engineers to build, salespeople to sell, marketers to acquire customers, operators to run the business and managers to coordinate the increasingly large number of people doing all of those things. More growth generally required more organizational capacity, and more organizational capacity generally required more capital.

AI is one of several technologies weakening that relationship. If a small team can access capabilities that previously required a much larger organization, the interesting result is not merely that the company has a lower burn rate. It may be able to test more ideas before committing resources, operate across more functions with fewer people, pursue markets that were previously too small to justify the organizational cost, or spend capital on constraints other than labor.

Calling this “efficiency” understates what may be happening. Efficiency implies that we are building essentially the same company for less money. The more interesting possibility is that we stop building the same company.

That is also why the distinction between incumbent adoption and new-company formation matters so much. If we evaluate AI primarily by looking at whether large companies have successfully reduced their existing costs, we may be looking for its most important effects in the wrong place. An incumbent has to unwind an organizational structure built under an earlier set of assumptions. A new company simply never has to create it.

We extended this argument in Portfolio Construction Is Changing. But So Is the Company Inside the Portfolio. Venture investors are already asking whether lower capital requirements should change fund construction, ownership targets, reserves and the number of companies a fund can support. But those questions ultimately depend on what happens to the economic unit being financed. If companies can reach meaningful scale with different headcount curves, different capital needs and different organizational structures, the implication is not simply that venture-backed companies become cheaper. It is that the shape of the venture-backed company itself may be changing.

We should be careful about overstating how far this goes. Software becoming cheaper does not make customers easier to acquire. AI does not eliminate regulation, distribution, trust, physical infrastructure, good management or the need to make something people actually want. New technological capabilities also create new costs, dependencies and risks. There is no serious reason yet to assume that every successful company will become a tiny team surrounded by agents.

But that uncertainty is an argument for better experimentation, not for reproducing the old organizational model by default.

What the AI backlash gets right

The most useful thing about the current skepticism toward corporate AI is that it restores the burden of proof. Companies should have to explain how AI is improving the business rather than pointing to the existence of an AI initiative. Executives should distinguish between productivity that has actually materialized and productivity they hope will materialize. Investors should be skeptical when ordinary restructuring is packaged as evidence of technological transformation. And technology leaders should be perfectly comfortable concluding, as Chili’s has, that fixing the Wi-Fi is a better investment than deploying another chatbot.

That is not a retreat from technological ambition. If anything, it demands more of innovation. It requires companies to distinguish between adopting a new technology and actually reconsidering what the technology makes possible.

The same standard should apply to founders building new companies, but the implications are different. They should not add AI because an AI-native company is supposed to contain a certain quantity of it. They should ask, with as little attachment to inherited assumptions as possible, what the best version of this company would look like if it were being designed for the technological environment that actually exists today.

Seen this way, the alternative to AI-washing is not AI skepticism; it is more rigorous company building.

For an incumbent, that may mean discovering that the most important technology investment has nothing to do with AI. It may mean using AI to make an existing operation substantially better. Or it may mean recognizing that a change in technological capability has created an opportunity large enough to warrant a new business altogether.

For a founder, the same discipline begins one step earlier. There is no organization to transform, no workflow to protect and no historical headcount model to justify. The opportunity is not merely to use new technology inside the company. It is to reconsider what the company needs to look like because that technology exists.

That is the part of the current AI debate we think deserves considerably more attention. The most interesting question is not whether companies are using enough AI. It is whether they are being honest enough about the problems they are trying to solve, and ambitious enough about what the answers might allow them to build.

‍