East-Digital Interview with Ishan Don in Leader Magazine (Eastern Switzerland): The Co-Pilot Answers. The Agent Takes Action

Many companies are experimenting with ChatGPT or Copilot. According to Ishan Don, however, the real transformation is only just beginning. In this interview, the founder and CEO of St. Gallen-based StackWorks AG explains why AI agents could take over entire business processes and why many companies have so far seen little real benefit from Gen AI.

Ishan Don, ChatGPT, and Copilot have now found their way into many companies. So why do you still refer to AI agents as a new category?

In short: A Copilot responds; an agent takes action. ChatGPT or Copilot generate text or code on demand. The human user continues to control the process. An agent, on the other hand, monitors conditions, makes decisions, and independently carries out actions—for example, in an email inbox, a CRM, or an ERP system. This moves us beyond the traditional “tool” paradigm. An agent doesn’t just write a suggestion—it completes the task.

"AI rarely fails because of the technology, but because of processes."

Where do Swiss SMEs currently stand in terms of AI adoption?

Right now, we see two types of companies. Some have long since adopted ChatGPT, Gemini, or Copilot and are now looking for the next step. Others are still watching and waiting. At the same time, developments are moving at a breakneck pace. Things that were considered innovations just a year ago already seem outdated in some cases.

In my view, 2026 is likely to mark the real turning point: a shift from “people using AI” to “processes using AI.” There’s another point to consider: Many companies have been using AI for quite some time, often without official approval. It is precisely this “shadow use” that is currently putting pressure on management to finally address the issue in a structured manner.

What specific tasks are AI agents already performing in companies today?

‍The range of possible applications has become very broad. We see uses in quote preparation, accounting, and IT security. I find its current use in first-level support particularly exciting. At a Swiss service company with about 100 employees, an agent now handles email triage, answers standard inquiries directly, forwards complex cases to the right team, and documents everything automatically.

In the past, the response time was two to three hours; today, it’s less than an hour. The team has remained the same size, but no longer spends the day sorting through emails; instead, it can once again focus on more complex cases.

Many companies are experimenting with GenAI but are seeing little real business impact. Why?

In my experience, there are usually three reasons for this. First, there is no clear process. Many companies implement AI tools without clearly defining the problem they actually want to solve. Second, there’s often a lack of a usable data foundation. If data is stored haphazardly or is incomplete, even AI can’t deliver consistent value. And third, many projects are approached from a purely technical perspective. However, AI rarely fails because of the technology itself, but almost always because of processes and responsibilities.

What risks or challenges are currently being underestimated when it comes to the use of AI agents?

Many companies are currently building agents without clearly defining what these agents are actually allowed to do. That is precisely where the risk begins, because it is often completely unclear what actions an agent is permitted to initiate on its own and on whose behalf it is acting. Data protection and data classification are also frequently underestimated. An agent cannot determine on its own which information is sensitive.

Then there’s the issue of auditability. When an agent approves orders or closes tickets, every step must remain traceable. In many cases, this means there still needs to be a “human in the loop”—that is, a person who remains actively involved.

"Many companies are building agents without clear guidelines."

You say that AI agents could become the dividing line between fast and slow companies by 2026. How exactly will that be evident?

This has a very direct impact on day-to-day operations. Companies that prepare quotes in hours rather than days win contracts faster. Those who organize support more efficiently—without constantly hiring additional staff—keep their margins stable.

And those who can test ideas or hypotheses within a few hours learn faster than the competition. As a result, AI is playing an increasingly important role in determining how quickly and efficiently companies operate.

What specific steps should CEOs take today if they want to address this issue effectively?

First, AI must become a top priority. We need someone in charge who has the authority, budget, and time. In my view, simply passing the issue off to IT won’t work. Many IT departments think primarily in terms of systems and tools.

However, AI is first and foremost about processes. Second, companies should start with two or three specific use cases where time or profit margins are currently being lost—such as in quoting, support, or accounting.

This is where initial projects can be tested in a targeted manner and experience can be gained. Without clear permissions, a well-organized database, and transparent processes, no AI solution will be able to scale in the long term .

Source: Published in Leader Magazine Eastern Switzerland, June/July 2026 issue ‍

Text: Patrick Stämpfli