I’ve spent a few weeks now inside the ESADE course — Rethinking Business With AI, and there’s a module on process automation with AI agents that has stirred me quite a bit.
Not because it revealed something completely new to me, but because it put order into something I had all mixed up. And when you have something muddled in your head, making good decisions is almost impossible. So I’m going to try to do the same for you.
Automation with agents is, today, one of the most powerful tools to multiply your work without multiplying your hours. But it has a huge problem: people treat it as if it were magic, when in reality it’s plumbing. And with plumbing, context is everything. Installing a pipe in a student flat is not the same as overhauling the water network of a hospital. Both are pipes. The mistakes don’t cost the same.
And what I’ve learned is that the process isn’t technical, it’s conceptual. There are two completely different worlds that share the same vocabulary and almost nothing else.
The Enterprise world
Here our plumber isn’t the neighbourhood guy who pops up to fix a dripping tap: he belongs to a team led by an architect and works within a construction project that demands planning and study.
When a large company automates with AI, it isn’t buying a tool: it’s making an architectural decision that will touch:
- Legacy systems.
- Regulated data flows.
- Impact on IT and Business teams.
- Process redesign.
- QA processes.
- Business validation processes.
Here an agent can’t just “try things out”. Every decision has real consequences, and the margin for error is narrow. This demands governance, traceability and human validation at the critical points. The biggest mistake in Enterprise environments isn’t technical, it’s about expectations: automation is hired as if it were an employee who shows up on Monday already knowing the whole context. And no. Skipping the definition and fine-tuning phase is like handing the warehouse keys to someone who just arrived on their first day.
Lightweight automation
The second world is completely different. These are lightweight automations. Our plumber is able, in this case, to understand what’s happening and, with his tools and almost without any prior analysis, fix that temperamental tap.
The small company that wants its CRM to update itself; the consultant who needs their follow-up emails to go out without having to remember; the team of five who want their weekly reports without manual intervention.
Here what works is speed, simplicity and pragmatism. You don’t need a complex architecture. You need a flow that works and that anyone on the team can understand even if they can’t code. The success criterion isn’t the same either: you don’t ask whether the system is auditable. You ask whether it has saved you two hours a week. If the answer is yes, it’s a success. Full stop.
When the worlds get mixed up
The problem comes when someone applies an Enterprise mindset to a small context or, worse still, when someone with a lightweight mindset lands in an Enterprise environment and pushes agents into production like someone laying pipe on a Sunday at a construction site to get ahead without coordinating with the rest of the team. However good their intentions, our plumber is going to cause us a problem.
AI agents are very obedient and very literal workers. They do exactly what they’re told, at a speed no human can match, and they make mistakes in ways no human would ever have imagined.
That’s why the design of the agent, its limits and its supervision points aren’t secondary details. They’re the core of everything. Smart automation doesn’t take work away from you: it changes your work. You stop doing repetitive tasks to design systems and make the decisions that truly require judgement. That sounds good. And it is, as long as you understand that this change doesn’t happen on its own. Someone has to design it, test it and keep it alive.
That’s plumbing too. And good plumbing is never noticed. You only notice it when it fails.
Where are you with automation? A small team chasing efficiency or an Enterprise environment navigating complexity?