The chatbot won't tell you to stop

Somewhere out there, right now, a CEO or founder is typing this into a chatbot:

“Hey Claude. Can you please help me create an AI strategy for my company? The business has been struggling so I want to release half of my employees and replace them with ChatGPT and Claude. Maybe also Grok, Elon made it. Help me create the strategy and the newsletter for the investors.”

We didn’t invent this little scenario to be mean. Prompts like this get typed every day, in nicer words, in worse words, in German and English and everything between. We have actually seen them, too, when leaders and founders share with us “the work they have already done”.

Struggling business, mounting pressure, and a machine that answers instantly, politely, and for free. Of course people ask. Why wouldn’t they? Tell you the truth, we did it too. And you probably, too. And here’s the uncomfortable part: the chatbot will help. It will produce a confident, well-structured strategy. The strategy might even be pretty good, and more than this: it will be confident. Phased rollout, projected savings, a risk section with reassuring bullet points. It will draft a genuinely elegant investor newsletter about “operational excellence through AI transformation.”

Not once will it say the one thing an actual expert would say within thirty seconds:

• “Stop. This plan will hurt you.”
• “Stop. This is not the way to go.”
• “Stop. Reconsider this whole endeavor.”

Why won’t the machine save you from yourself?

It’s not because the models are bad. Or because AI is bad all by itself. We run an AI-adjacent business; we use these tools daily and happily and help others use them, too. It’s because of what a language model fundamentally is: an extremely capable assistant that works with what you give it and is trained to be helpful about it. It will not tell you that your premises are flawed or just simply incorrect.

It doesn’t know that your “struggling business” is struggling because of a pricing problem, not a headcount problem. It doesn’t know that the twelve people you’re about to release hold most of your customer relationships and all of your undocumented process knowledge. It doesn’t know that your investors will read “we replaced half our staff with chatbots” not as innovation but as a distress flare. It will accept your premises as true and factual: Of course your product is good because you told AI so. Of course you are competent because you told AI so. It doesn’t know your biases.

It doesn’t know, because it can’t know. It has never sat in your business (or any business, really. It read about them). And crucially: when the plan fails, nothing happens to the chatbot. Something very much happens to you.

One of our experts, Laura McGillicuddy, put the dividing line better than we ever have: “For me, the line is consequence.” The greater the impact of a decision on real people, the more you need human judgement, context and accountability. Now ask yourself what carries more consequence than a plan that decides who keeps their job.

By her own standard, and by ours: an AI strategy is roughly the highest-consequence decision most companies will make this decade. Which makes it precisely the decision you should not delegate to a system that agrees with you by design.

What the real version looks like

So what would an actual expert tell our CEO? Probably something like this, and notice how it starts:

Not with the technology. With the problem. “The business has been struggling” is doing a lot of work in that prompt, and nobody has asked why. Another of our experts, Michel Knecht, calls this the most common leadership mistake in AI: treating it as a tool question. “Which tool should we buy?” when the real question is which work should get better, faster or lighter.

Not with replacement. With relief. The unglamorous truth from everyone who does this work for a living: the reliable value of AI today is not removing people. It’s removing friction from people. Better first drafts, faster orientation in documents, meeting minutes that write themselves. Michel again, from his work with Swiss administrations: the win wasn’t automation, it was employees saying “this actually helps me in my work.” That’s what real adoption sounds like. It doesn’t sound like a severance package.

Not alone. With the people who do the work. Laura’s advice for the “where do we start” question: start with your employees. They already know where the time goes and where the friction lives. A strategy written by a chatbot at midnight has none of that knowledge in it. A strategy built in one honest hour with your team has all of it.

And with a name attached. Every AI use case needs a human owner. Someone who checks the output, answers for the mistake, and can pull the plug. A chatbot can’t be that person. Neither, by the way, can “IT.” Again, not because they’re incompetent — but because they’re the wrong person to ask.

The punchline you saw coming

Yes, we see the irony. We’re a company that believes in asking humans, telling you not to ask a machine, in a blog post you’re reading on the internet. On how to use AI best. Maybe you even asked chatGPT for a summary of the post. Fair.

But notice what actually happened in that opening prompt. The CEO didn’t need a document. The CEO needed someone to push back. That is the one thing the chatbot will not and can not do, and the first thing a real expert does. Pushback is not a bug of human advice. It’s the product and the most ideal outcome.

Use AI for the drafts, the summaries, the analysis, the 2 a.m. questions. Genuinely. It’s superb at that and hiring an expert might be a complete waste of your money. But the decision about what AI should do to your company, your people and your name? That one deserves a counterpart who has done it before, knows what it costs when it goes wrong, and is accountable for the answer.

You can have that conversation this week. It takes an hour.

Just ask. A person, this time.