Raimo Seero, Uptime CTO
When the first tractors appeared in the fields, their power was measured in horsepower. When photography began reaching the masses, it was judged by how accurately it could imitate painting. In business today, artificial intelligence is often being approached in much the same way: people try to understand a new technology through the logic of the old world. Instead of asking what new outcomes it makes possible, they focus on how to make AI do existing work in the most human-like way possible.
New technology is almost always interpreted through the logic of the old world. That is only natural, because the human brain looks for familiar patterns and tries to translate new phenomena into something it already understands. But that same reflex is currently one of the biggest factors limiting how companies apply AI in business.
Most organisations essentially want the same thing from artificial intelligence: to do existing work in the same way a human would. They expect AI to follow a process, take the correct steps, and behave according to a script written for people. But that is the wrong question – and the wrong question will always produce the right answer to the wrong problem.
The real question is not how well AI follows a process, but what outcome we actually want to achieve. That is where the dividing line lies between two fundamentally different approaches: on one side is the desire to make AI serve the existing way of working, only faster; on the other is the willingness to ask whether the current process is even necessary in the form we know it today.
Software Development: From Code to Outcome
This is especially clear in software development. The typical picture today looks like this: a developer writes a specification, AI generates blocks of code, the developer reviews them, makes corrections, and tests them. In this model, AI essentially functions as a faster autocomplete tool – the process itself remains the same, only one step becomes quicker.
Take a concrete example: building a payment solution for an e-commerce store. The old way is that a developer writes the payment form, manually tests five scenarios, sends it to QA, and the cycle repeats. But where we really want to get to is an outcome-oriented approach, and that looks different: AI generates the payment form, automatically creates 200 test scenarios, simulates failed payments, expired cards, and network issues, and returns to the developer only the three cases that require human judgment.
At the same time, we should not forget to evaluate whether a payment solution is even an area where AI should be allowed to operate freely. AI should not be deployed everywhere simply because it can do something faster. First, we need to assess the risks associated with a particular solution and whether those risks can be mitigated sufficiently.
Just as we would not trust every person with every job, we cannot trust AI equally in every process. But that does not mean AI’s use cases are limited. On the contrary, in many cases, a small process change and well-designed control mechanisms are enough to manage the risks and unlock significant value from AI.
As a result, steps that once seemed unavoidable may turn out to be unnecessary – because they mainly existed to support human work, while AI may not need them at all.
The Issue Is Not Technology, but Leadership and Risk Management
This is the heart of the matter. If we optimise a process, we get faster processes, which is valuable – but still a limited gain. If, however, we start from the desired outcome, we often discover that some steps are not essential at all; they simply reflect the way humans have historically approached the task.
That is why implementing AI is not primarily a technical question, but a leadership question. The next time an organisation discusses using AI in a process, the first question should be: what outcome is this process meant to produce? Then it should honestly assess whether the current path to that outcome is the only possible solution – or simply the one people have become used to.
AI is not just a new tool for an old model. It is a reason to ask whether the old model is necessary at all. Companies that are willing to genuinely ask that question will not just optimise existing processes – they will build entirely new ones.