For the past two years, we’ve talked a lot about productivity. Productivity of programmers , lawyers , doctors , consultants , financiers , teachers , and even students . The promise was simple: give me an artificial intelligence tool and I’ll do more in less time.
It’s true that an employee with access to a good generative model can type faster, summarize better, program with assistance, or prepare reports in less time. However, many companies are discovering an uncomfortable truth: improving individual productivity doesn’t always improve organizational efficiency .
Using a lot of AI doesn’t necessarily mean more profits for the company.
A company can have hundreds of employees using AI every day and still not have changed its processes. It can pay for licenses, APIs (Application Programming Interfaces: a set of rules that allows different software applications to communicate with each other), and pilot projects, and still not know what it’s gaining. It can have a lot of activity and many demonstrations… but little impact on the bottom line.
That’s where the change begins. The major AI labs are realizing that the business isn’t just about selling models, but about making those models generate value within companies.
In this context, on the other hand, generative AI has a powerful but uneven characteristic: it amplifies the capabilities of those who know how to use it . However, for someone without criteria, method, or context, it can become noise, dependency, or false productivity.
This same effect applies to businesses. Having a subscription doesn’t transform an organization. Having a chatbot doesn’t automatically improve the user experience. For that to happen, you need business knowledge, process integration, organizational redesign, impact measurement, and execution capabilities.
That’s why the business model of large platforms like OpenAI is starting to change. The first stage of generative AI was based on charging for access: a monthly subscription, a price per user, a cost per token —a pricing model used in artificial intelligence and language models (like GPT-4 , Claude , or Llama ), where payment is exact and variable based on the amount of text processed. It made sense: the technology was new, and companies wanted to experiment. The question was: “What can we do with this?”
Now the question is different: “What business problem does it solve, and how much value does it generate?” The company doesn’t want to pay simply to consume tokens —the basic text components that language models process. It wants to pay to reduce costs, increase revenue, accelerate processes, improve service, or decrease errors. It’s not looking for “an advanced linguistic model,” but rather a selection solution that reduces screening time, a 24/7 support system, or an assistant that prepares personalized sales proposals.
The economic unit is no longer the token , but the result
In today’s offices, AI is no longer an occasional feature but has become an everyday part of the workflow. While the cost per token may decrease, the total cost can increase with more complex queries and greater automation. Furthermore, many companies want to invest in AI but don’t always know how to generate a return on that investment . Moreover, there is an ever-growing number of models, tools, and vendors to choose from. What seemed extraordinary yesterday may be standard functionality tomorrow.
In this scenario, generative AI systems are progressively becoming a commodity : an essential component, but increasingly less of a differentiator on its own . Their added value no longer lies solely in the model’s intelligence, but in its ability to integrate into an organization’s operational reality: its data, processes, legal constraints, legacy systems, culture, and metrics.
The value goes from the laboratory to the real world
For these reasons, the technology provider is beginning to transform into an operational partner . The former sells access to a tool. The latter commits to a result: it talks about response times, operating costs, customer satisfaction, error reduction, regulatory compliance, and return on investment.
This leap brings tech giants closer to the realm of business consulting, because enterprise adoption of AI is no longer a purely technical problem, but a matter of organizational transformation. It requires identifying use cases, prioritizing them, integrating them, measuring them, governing them, and scaling them. It’s essential to know where AI creates value and where it only creates the appearance of modernity.
And that’s not easy. AI in a demo is usually clean, but in a business it’s often messy . Data isn’t where it should be. Systems don’t always communicate with each other. Processes have exceptions. Regulations impose limits. Success depends not only on the model responding well, but on the entire system working.
The product logic also changes. An AI product can be sold to many companies with minor variations. But a solution that transforms a process needs to be tailored to the sector, the company’s size, its digital maturity, its data, and its objectives. The more specific the solution, the greater the perceived value.
This change also affects professional profiles. Alongside programmers, those who transform models into real solutions will gain prominence: people capable of communicating with a technical team and an operations manager, designing a workflow, measuring results, and adjusting the system to the client’s needs.
Tangible solutions for businesses
The great paradox is that AI labs have created a horizontal technology, but they need to capture vertical value. A machine learning model can be used for almost anything. But a company only truly pays when that “almost anything” translates into “this concrete thing that improves my business.”
If generative artificial intelligence started out being sold as a tool to do more things faster, its business maturity will depend on a much more demanding promise: making organizations work better.
Perhaps that’s why the future of AI labs doesn’t resemble that of a traditional software company so much as a complex mix of infrastructure, consulting, integration, and operations. OpenAI and Anthropic have already begun this transformation. In this shift from technology provider to operational partner , the true business model for artificial intelligence may be emerging.
Author Bio: Antonio Pita Lozano is Professor of Data Science and Artificial Intelligence at UOC – Universitat Oberta de Catalunya
