AI Can Write Your Emails. But Should It?

1. Notice when AI is forming your thoughts, not just supporting them. Using AI to draft a message, adjust your tone, or frame an argument is different from using it to summarize a document or find a bug. The first is worth doing consciously, not by default.
2. Map what you delegate to AI against what you want to keep sharp. Calculators took arithmetic — and that trade-off was fine. Communication and judgment are different, and you shouldn’t just let these skills slowly degrade.
3. Know where accountability sits before you deploy anything. The deployer of an AI system is the sole responsible party for its decisions. Be careful about who or what you delegate your responsibility to.
4. Invest in understanding business processes, not specific tools. English is the new programming language — but only if you can articulate what you actually want the system to do. Process knowledge and technical understanding is what makes a prompt useful.
5. Build in time for upskilling before the gap becomes visible. The tools that matter in two years may not exist yet. A habit of regular, low-friction learning compounds faster than any single course or certification.

The Advanced Analytics tribe was set up years ago to bring machine learning into the bank's decision-making — moving away from rule-based systems toward models that could work with data at scale. Somewhere in the middle, generative AI arrived and changed the scope of what was possible. The team now runs hybrid models, combining classical machine learning with what large language models and agents can do.

This depth of experience makes Renato somebody who can openly speak not only about the benefits of AI adoption, but also about the downsides – and one of them is a risk to lose our own agency and cognitive abilities. He is describing a pattern visible across society: "When was the last time you wrote a message without using AI — without adjusting the tone, without running it through something?"

As Renato says, we delegated arithmetic tasks to calculators decades ago, and most people can no longer do complex mental calculations — and that has been largely fine. The cognitive cost was acceptable because arithmetic, while useful, is not the thing that defines how we think and relate to each other.

"Now we are dealing with a different animal. It is mastering our discourse, our communication — the very basic tool that defines civilization, which is our ability to communicate." When AI produces an idea for you, it is not saving you from a mechanical task but doing the thing that is most distinctively human.

"The more we get used to it, the more we abide to models, biases, and an ethos which is not ours," he says. That does not mean the answer is to refuse the tools, but it means the use has to be deliberate. This awareness is foundational for RBI's approach to AI: the education programs, the governance framework, the design of how AI systems are deployed. It's all about responsible AI.

Documenting code: highly automatable. Generating personalized content streams for business experts: highly automatable. Finding bugs in software: ready for automation. These are tasks where the output can be verified, where errors are recoverable, and the cost of a mistake is bounded.

Portfolio analysis, strategy development, anything involving judgment sits on the other side, defined by a thing that cannot be delegated: accountability. "Our legislation says that the deployers of AI systems are the sole responsible for the decisions that they are making," Renato says. AI in the co-pilot seat, human in the driving seat: it is the legal and ethical structure the work operates within.

The question of which skills to build feels urgent and even disorienting. The tools change fast enough that specific technical knowledge can become peripheral within a few years. Renato's answer is not to name a skill but to reframe the question.

"The separation between business skills and tech skills is getting more and more blurred," he says. Agents are now generating the coding infrastructure for most projects; the role of programmers has shifted toward supervising that work rather than producing lines of code directly. Which means the thing that actually matters is understanding your processes — well enough to translate them into what you want a system to do.

"English is the new programming language," Renato says. If you can articulate how a process works and what the inputs and outputs are, you can direct the tools to some extent. What you cannot replace with a prompt is the process knowledge itself, the understanding of tech logic, and the judgment about what the right outcome actually looks like.

The practical implication is not to pick a specific technology to master but to reserve time for constant, daily upskilling — and to be skeptical of any skill framed as the essential one for the next five years. Your curious attitude, smart habits, and flexible mindset are what will stay with you no matter what.