Can AI be a company’s competitive advantage?

Not as such: a tool your competitor can buy tomorrow is no one’s lasting advantage. Competitive advantage comes from how fast people learn to apply AI in everyday work. And learning speed is a result of leadership, not a procurement.

At Astu Labs we support leadership teams through AI transformation. We do not deliver technology — we help lead the change around it. From this vantage point we see the same pattern regardless of organization, industry, or size. Tools are acquired at roughly the same pace, from the same vendors, with the same promises. Agents are deployed to support the same routine tasks. And yet the results differ drastically. Same technology, different outcome. The difference cannot be explained by the technology, because everyone has the same technology.

Nearly every organization starts with the same thought: our industry is a bit special. From the perspective of AI adoption, that is rarely true. So far the challenges have been the same everywhere: where to start and what needs to be taken into account, which tools may be used and for what, whether the skills are sufficient, how to manage shadow use, and how all this shows up on the bottom line and in customer value.

If the challenges and the benefits are the same for everyone, then acquiring the technology cannot be anyone’s competitive advantage. The advantage comes instead from how well and how fast your people learn, and from how capably they apply what they have learned in everyday work. In other words: competitive advantage should be examined through what your people will keep doing going forward — in an era when cognitive work can be scaled without growing payroll costs.

Why isn’t an AI tool a competitive advantage?

An AI tool is not a competitive advantage, because a competitor can acquire the same tool at any time. This is not a new phenomenon but one of the oldest patterns in the history of technology. Electricity was once a revolutionary advantage for the factories that got it first, but once every factory was electrified, electricity became a mere cost item and competition was again decided by other factors. Nicholas Carr crystallized the same point about IT back in 2003: technology that everyone can buy is necessary but not differentiating.

Strategy research puts it even more precisely. A sustained competitive advantage can only arise from a resource that is valuable, rare, and hard to imitate (Barney 1991). A language model or a tool can be valuable, but it is neither rare nor hard to imitate. A competitor can make the same purchase at any time. The same test fells most AI agents too. If anyone can buy or build it, it is no one’s lasting competitive advantage.

What does research say about AI’s productivity gains?

Research shows that AI’s productivity gains arise from people and leadership, not from the tool itself. In MIT’s extensive report (NANDA 2025), roughly 95 percent of enterprise generative AI pilots produced no measurable impact on the bottom line. The study has drawn plenty of criticism, but the essential point is not the number — it is the reason. According to the report, the failures were not caused by the quality of the language models, by security, or by regulation, but by a learning gap. That is, by how to fit AI into the right workflows, structures, and ways of working. The technology worked, but the organizations did not learn to use it.

That the difference lies in what people do also shows in how the benefit is distributed. Brynjolfsson et al. followed a good five thousand customer service agents (QJE 2025) and found that productivity rose by about 15 percent on average, but by as much as a third for the least experienced. Same tool, different result for different people. The benefit does not follow the technology but who uses it, for what, and how they are led.

It is a matter of judgment, not of use itself. AI is not evenly skilled or unskilled. Its competence resembles a jagged coastline more than a straight border. In places AI reaches surprisingly far; in places it fails exactly where you would expect it to perform best.

This is precisely what makes the phenomenon treacherous. When researchers from Harvard and BCG (Dell’Acqua et al. 2023) gave consultants access to AI, performance clearly improved on tasks that fell within its zone of capability. On tasks that fell outside AI’s zone of capability, by contrast, performance actually deteriorated. The consultants trusted convincing but flawed AI-generated output.

How do you lead learning speed?

Leading learning speed belongs to the leadership team, not to the IT department alone, because above all it is about leading learning. Learning is not a technical variable — it is a result of leadership. People learn fast where they are truly met, where experimenting is safe, and where failure can be talked about out loud.

When it comes to building trust, adapting, and learning, servant leadership, for example, has exceptionally strong research evidence behind it (e.g. Lee et al. 2020, a meta-analysis of 130 studies). Technology is acquired once, but learning capability is built — and led — continuously, through everyday managerial work.

“What about our own data? Isn’t that exactly the rare resource?”

Many consider their data and its use their competitive advantage. It is a good argument, and we take it seriously. Proprietary data and integrations built deep into processes are the closest thing to a resource that passes Barney’s test described above — they can be valuable, rare, and slow to copy. But it is worth recognizing what their value, too, ultimately rests on. Data produces nothing until someone changes the workflow where it is used and knows how to judge when the model’s output can be trusted.

Barney himself names socially complex resources as the hardest to imitate: culture, trust, and the ways people work together. So the advantage data brings is likewise realized only through people, and the competition is decided by who learns to put their own data to use faster.

Three questions worth asking right away

We encourage leadership teams to test their own situation with three questions:

1. Are we measuring adoption or learning? The number of pilots and deployed agents is a vanity metric. Instead, align on and communicate what concrete change you are aiming for. Then assess whether the way work gets done has concretely changed over, say, the past three months. If the answer is no, why would the desired change happen later either?

2. If a competitor acquired exactly the same tools and agents tomorrow, what would we have left? What remains is your competitive advantage. If the answer is “not much,” you do not have one yet.

3. Are we leading people’s learning as deliberately as we manage technology purchases? Most are not. And that is precisely where the opportunity to stand out lies.

The question “how do we get competitive advantage from AI” ultimately turns into a question about people — skills, workflows, cost structure, and governance — and above all about how fast people learn. Learning speed is not bought with a license or an agent. It is built, and it is led.

Sources

• Barney, J. (1991). Firm Resources and Sustained Competitive Advantage. Journal of Management, 17(1), 99–120. 

• Carr, N. (2003). IT Doesn’t Matter. Harvard Business Review, May 2003. 

• MIT NANDA (2025). The GenAI Divide: State of AI in Business 2025. (95% of pilots showed no measurable impact on results; shadow use at ~40% / over 90% of companies.) 

• Dell’Acqua, F. et al. (2023). Navigating the Jagged Technological Frontier. Harvard Business School Working Paper 24-013. (Published in peer-reviewed form: Organization Science, 2026.) 

• Brynjolfsson, E., Li, D. & Raymond, L. (2025). Generative AI at Work. The Quarterly Journal of Economics, 140(2), 889–942. 

• Lee, A., Lyubovnikova, J., Tian, A. W. & Knight, C. (2020). Servant leadership: A meta-analytic examination of incremental contribution, moderation, and mediation. Journal of Occupational and Organizational Psychology, 93(1), 1–44.