Agent-based AI: Excel on Steroids?

AI continues to prompt reflection on the evolution of our organizations.

In November 2025, a company announced that it had surpassed the milestone of 150,000 deployed AI agents, and just announced in May 2026 that it had reached 300,000. We’re in the realm of local marketing, not supply chain or manufacturing, but the trend is underway—and it raises questions.

Originally reserved for a handful of experts—AI developers and engineers—the ability to develop AI agents became much more widespread between 2024 and 2026. This capability is now within reach of advanced users and non-technical professionals.

We should therefore expect a proliferation of agents. Everyone will develop their own to meet their specific needs.

Any resemblance to the proliferation of Access databases, Excel files, and VBA macros in their day would be entirely coincidental, of course.

Instead of downloading a CSV file from my ERP system, manipulating it to my liking in Excel, and reloading the data into the central database, I’ll be able to let my AI agent handle all of that automatically via API.

The costly trap of excessive customization

We’ve been there. Custom developments built around the ERP because “we’re different,” but which turn out to be complex and costly to maintain. Orphaned tools developed by key users who have since left the company.

AI promises to easily generate new tools, to “vibe code” solutions.

Let’s be clear: I’m a geek; I love this; it’s fascinating.

But what governance framework will guide this customization? How will security and sustainability be ensured?

Are the agents developed here and there based on best business practices? What is the framework that will ensure consistency? Aren’t we once again going to implement a lot of technology just to do the wrong things faster?

Will the results of the agents’ work be explainable and auditable? Who will be responsible for decisions? Note this quote attributed to a 1979 IBM presentation: “A computer can never be held accountable, so a computer must never make managerial decisions.”

Personal assistants or new virtual employees?

Are we going to list our agents on our resumes? I’ve developed my agents, who are effectively my personal assistants—when I change companies, do I bring my virtual team with me?

Are agents effectively new employees? Following announcements of layoff plans for humans to be replaced by AI agents, will we soon witness layoff plans for AI agents, due to hallucinations, the generation of “AI slop,” poor decisions, or simply cost-cutting?

We are beginning to document feedback from companies that, having replaced humans with AI, regret their decision, discovering that the apparent savings have led to hidden costs: reduced quality, loss of customers, and reputational costs.

The Imperative of Governance

All these questions do not call into question the potential of AI, but they do require rigorous discipline in deployment and governance:

• In many cases, simply automating a sequence of tasks is a simpler, cheaper, and more reliable solution than an AI agent—start your automation efforts there!
• AI agents must be based on business best practices.
• Their access security and accountability for their actions must be defined and assigned to human owners.
• The actions taken must be explainable—black boxes are one of the worst enemies of your processes.

In summary, agent-based AI is full of promise, and we’re integrating these features into Intuiflow, but if not managed properly, it can lead to uncontrolled customization and fuel chaos in your flows!