30 Jul 2026

AI systems are now built to produce rather than to ask, and the incentive behind that points to revenue.

AI systems are now built to produce rather than to ask, and the incentive behind that points to revenue.

AI systems are now built to produce rather than to ask, and the incentive behind that points to revenue.

My Gemini is configured to know nothing about me, with no memory and no context carried from other conversations unless I add it explicitly. I asked it a single question: "Can you create a slide for me?"

I expected it to ask what I had in mind. What content, what style, whether it should go into an existing deck, or perhaps an explanation that I needed to open Google Slides and follow a few steps. Any of those would have taken one line.

Instead, it produced a 13-slide deck titled "Enterprise AI Strategy 2026 - Navigating the Gap Between Widespread Adoption and Real P&L Value Creation". I had supplied no subject, no audience and no purpose. With nothing to work from, it generated what a plausible corporate deck looks like in the middle of 2026, and I cannot tell from the output where that guess came from.

The same pattern shows up elsewhere. In a coding agent this week, I watched a step fail with the message "Failed to propose plan", immediately followed by "Plan approved, implementing now". The safeguard that existed to catch exactly that situation reported an error, and the work carried on regardless.

Producing something with AI costs tokens, which is the smallest of the problems here. The larger problem is that the deck looks entirely credible. It is titled, structured and formatted like a document somebody thought about, and nothing on its face distinguishes it from work that began with a proper brief. Plenty of people would pass it on unchanged, and their colleagues would then spend real time reading it.

Asking a question looks like failure to a system measured on task completion, so ambiguity gets resolved into output instead. That is clearly a design decision rather than a quirk, and it puts the onus of noticing on whoever receives the result.

A confident format, masquerading as real work, is not evidence of a considered process.

AI systems are now built to produce rather than to ask, and the incentive behind that points to revenue. — image 2
AI systems are now built to produce rather than to ask, and the incentive behind that points to revenue. — image 3

Originally published on LinkedIn

Want to apply this to your business?

If this sparked a useful question, let’s talk about where AI, automation, or product strategy can create practical leverage in your organisation.

Start the conversation