The third slide already promises to “radically transform” a profession. The fifth introduces an autonomous agent. Someone in the room checks their phone. The problem is not necessarily the product: it is that its story sounds like all the others. For entrepreneurs, AI fatigue presents a challenge of communication as much as credibility. The pitch that stands out no longer announces a revolution. It shows a specific difficulty, provides evidence and explains where the machine stops.
This analysis looks ahead to September 2026, drawing on events known through December 2025. The developments envisaged for 2026 are therefore forward-looking assessments, not an established account of events.
When promises become background noise
Since ChatGPT reached the general public in late 2022, the language of artificial intelligence has permeated sales presentations. Copilots, assistants, then agents: each generation promises to take on more tasks. Microsoft, Google, Salesforce and numerous software vendors have integrated these features into their offerings. For a startup, displaying “AI-powered” is therefore no longer enough to establish a clear distinction.
This normalization meets another reality: buyers have learned to ask less sensational questions. Who checks the answer? What data leaves the company? How much does integration cost? What happens when a document is missing? Weariness does not necessarily mean rejection of the technology. It may reflect evaluation fatigue: too many promises to assess, too many demonstrations that are difficult to compare, too much work hidden behind the apparent simplicity.
Klarna’s story illustrates the tensions in this narrative. In 2024, the company heavily promoted the performance of its customer service assistant. In 2025, its chief executive acknowledged that an excessive focus on cutting costs could undermine quality, while reaffirming the importance of access to human representatives. The lesson is not that automation always fails. It is that a narrative focused solely on replacement becomes fragile as soon as real-world experience enters the conversation.
Start with a scene, not a market
A good pitch starts where the customer loses something: time, a sale, attention, trust. Take a hypothetical example. At a small or medium-sized manufacturing company, a sales manager receives requests for quotes accompanied by drawings, emails and old references. Before she can even work out a price, she must track down the information, identify missing items and follow up with colleagues. That is a scene worth telling. “We are reinventing industrial productivity” is not.
This precision draws on an essential interpersonal skill: listening. The founder must be able to describe the problem in the other person’s words, without immediately correcting them with technical terminology. It is also important to distinguish between the user, the buyer and the person who will bear the risk. The sales manager wants to move faster; the IT department wants to control access; the chief executive wants to understand the expected return. A single narrative can connect them, but it cannot erase their differences.
Evidence must stand up to questions
After the problem comes the evidence. Not necessarily an impressive figure: an observable result, accompanied by the conditions under which it was achieved. A prepared demonstration shows that the product can work. A trial using representative cases begins to show under what circumstances it works. Repeated use makes it possible to examine whether the benefit persists. These three levels are not interchangeable, even when they produce the same appealing video.
Research on AI at work calls for precisely this caution. Studies published before 2026 observed gains in certain writing, support and programming tasks, with results varying according to users’ experience and the nature of the work. Other experiments showed that assistance could mislead users when a task exceeded the system’s capabilities. A gain in one specific setting therefore cannot be turned into a universal promise.
For our hypothetical company, the evidence could take the form of a test comparing case preparation with and without assistance. The types of requests would need to be specified, verification time counted and errors reported. If the test is still too limited, it is better to say so. A pitch becomes more robust when it distinguishes between what has been measured, what has been observed and what remains a hypothesis.
The limit is not a footnote
Entrepreneurs often fear that a caveat will break their momentum. A well-articulated limit can, on the contrary, make the proposition understandable. The tool prepares a case file, but does not approve the price. It suggests a response, but does not settle a dispute. It sorts documents, but requests intervention when essential information is missing. This boundary allows the customer to visualize responsibility, not just capability.
That still requires avoiding the lazy formula: “A human remains in the loop.” Which human? At what point? With what information, and how much time to check? Constant approval requirements can consume the promised gain. Token approval can let errors slip through. In a pitch, explaining the handover mechanism matters more than saying the word “oversight.”
Three steps to change how you persuade
This shift requires less rhetorical flair than discipline. Before a meeting, the team can rework its narrative around three simple steps:
- Describe before you characterize. Replace “revolutionary” with a workplace situation, a point of friction and its consequence.
- Show before you generalize. Present a documented result, its scope and the effort required to achieve it.
- Define boundaries before you reassure. Identify a failure scenario, how to detect it and the fallback solution.
Your approach matters as much as the wording. When faced with an objection, restating the concern before answering shows that you are trying to understand rather than win a duel. Being able to say “we have not tested that yet” avoids turning every question into a sales commitment that must later be honored. And acknowledging that a conventional process is sometimes sufficient can make the case where AI genuinely adds value more credible.
Sell an improvement, not a belief
Looking ahead to September 2026, it is reasonable to expect that the proliferation of offerings will make this restraint more valuable. It guarantees neither funding nor a signed deal. But it makes a concrete decision easier: to run a trial within a defined scope, with success criteria and the option to stop. The pitch then becomes the beginning of a framework for trust, rather than a contest of enthusiasm.
What now? If technical capabilities continue to advance, the temptation to turn up the volume on promises will be strong. Entrepreneurs might benefit from doing the opposite: explaining more precisely what changes, for whom and under what conditions. The decisive skill may not be appearing visionary, but making an ambition verifiable. In a room weary of AI, that is already one way to recapture attention.
