Yesterday I experienced, together with Nacho, our CEO, one of those moments that reorient your professional compass. We were facing a wall: a financial report of extreme complexity necessary to apply for a new project. It wasn't just filling cells; it was a labyrinth of revenue annualization, inflation-indexed expense projections, depreciation and tax calculations.
Let's be honest: neither Nacho nor I are career financiers. Without external help, completing that document with the necessary rigor would have taken us at least three days of intensive work and much uncertainty. We knew that either we hired a full-time financier, or we ran the risk of not being up to the project.
We decided to change strategy. We didn't ask the AI to "do the work". We asked Claude to be our senior strategy consultant.
The paradigm shift: From "do" to "ask"
Through prompt engineering, we designed a work process where the AI was not a simple executor, but a conductor. We asked it to initiate a step-by-step consulting process: it had to request the necessary information from us, validate each data point, perform the calculations and, most importantly, document the reasoning behind each figure.
And then the "click moment" happened.
Instead of rushing to fill the file with invented data to try to please us (the dreaded hallucination), the model stopped. It analyzed the inflation and depreciation variables and threw us a technical question to get the exact context before continuing.
"This really changes the game," Nacho commented at that moment.
Why is this fact differential?
As experts, we know that the biggest risk of AI is its complacency bias. However, when you design a flow where the AI has permission to stop and request context, the quality of the result skyrockets:
Long-term reasoning: The model maintained coherence throughout the session, connecting asset depreciation with the final cash flow flawlessly.
"Senior Expert" Quality: Seeing the assumptions that the model proposed, we realized that the technical level was superior to what we ourselves would have achieved. We had closed the financial talent gap in one afternoon.
Defense capability: The most powerful thing is not just having the finished document. Thanks to the model documenting the entire process, we can now present it and, above all, defend it before any auditor, because we understand the "why" behind each number.
From 3 days to 90 minutes
What would have previously required weeks of back and forth with external consultants or days of internal blockage, we solved in an hour and a half.
This real case shows us that AI is not a tool to write emails faster; it is a reasoning engine that, well directed by profiles like Nacho or me, allows a team to execute tasks of a technical complexity that was previously out of their reach.
Yesterday's great lesson is clear: AI is not going to replace experts, but it will make those who know how to iterate with it indistinguishable from a team of experts.