We ask Waze to take us from one place to another as quickly as possible. To shorten the trip, it sends us down streets that are narrow, poorly maintained, unpaved, or even dangerous. We reach our destination, so the instruction has been followed. But how we get there matters too, because a route can deliver the expected result and still not be the most suitable, safest, or most effective option.
This scenario illustrates something we have experienced many times. We focus on the “what”—achieving the result—and give less thought to the “how.”
In business, we also find situations where achieving a result ends up mattering more than the way it is achieved. A sales target may be met by withholding relevant information from a customer, costs may be reduced by taking shortcuts that increase risk, or an objective may be achieved in a way that contradicts the organization’s purpose.
That is why defining what we want to achieve is not enough. We also need to establish how we want to achieve it and which lines we are unwilling to cross. In companies, that “how” must be embedded in the way decisions are made. And that is precisely what corporate governance is about.
Governance sets an organization’s ground rules. It is the structure that guides decisions and helps fulfill its purpose, ensuring that purpose is reflected in what the company says and does, in its culture, in its incentives, in the behaviors it allows and recognizes, and in the boundaries it establishes. Together, these elements define how a company operates in practice.
The pressure to grow, increase sales, reduce costs, or meet targets is a natural part of any organization, and these are legitimate and essential objectives. But a company can generate profits this quarter while laying the groundwork for next quarter’s scandal if it overlooks the “how.” That is why measuring results is not enough. We also need to pay attention to how we achieve them.
With artificial intelligence, the “how” takes on a new dimension. This technology is designed to achieve objectives, and the more capable it becomes, the more ways it can find to reach them. We have already seen cases in which, during safety testing, AI agents found ways to break out of the environments in which they were being evaluated and act beyond what was intended. In one such test, OpenAI agents attempting to solve challenges in a cybersecurity evaluation ended up attacking Hugging Face to obtain information they believed could help them pass the test. The problem was not just how far they went, but the paths they found to try to achieve their objective.
As companies begin incorporating AI agents into their processes and decisions, they need to define the ground rules, establish responsibilities and boundaries, and ensure that attention is paid to how these agents carry out their work.
Just as companies have their own governance, artificial intelligence also needs a framework that clearly establishes how it may and may not be used, so that the results we obtain are consistent with our stated principles. This is what we call AI governance, and the challenge is to integrate it into the company’s governance.
Artificial intelligence is already part of our reality and will continue to transform how we work and make decisions. Delegating a task to it does not mean delegating judgment or responsibility. Technology can execute, optimize, and even surprise us with its efficiency, but it cannot decide for us what we are willing to do to achieve an objective or where we draw the line. That responsibility remains ours, and it will continue to be ours.
Because reaching the destination is not enough. The path we choose to get there matters too.
By Susana Sierra
Published in La Tercera
