In previous columns, we saw how AI multiplies existing capabilities and why “AI + me” is now the unit of work. Yet a puzzle remains: if AI merely amplifies what we bring, why do so many fail to activate the multiplier? Because with generative AI, the user, not the system, must supply the manual.
Old technologies came with manuals for a reason. Their behavior was predictable. Functions were predefined. If steps were followed, outcomes were repeatable. Complexity was absorbed by design. The burden of thinking sat with the system, not the user.
That assumption no longer holds.
Generative AI does not behave like a tool with buttons, but like a capability that responds to intent. It does not know what you want unless you describe it, nor what success looks like unless you define it. The work once hidden inside product design has shifted back to us.
This is why so many AI outputs look impressive but fail in practice. The language is fluent. The structure is complete. Yet the result misses the point. The problem is often not the model; the task was never properly specified.
When asked to “help write a report,” a human would ask follow-up questions: for whom, for what purpose, under what constraints. AI does not. It must decide what the report is for. Is it meant to inform, persuade, justify, or deflect blame? Lacking guidance, it picks one, delivers immediately, and fills the gaps with assumptions. The result often reads well and quietly fails, because it optimizes for a purpose the user never articulated.
AI operates with multiple paths and no default definition of “right.” It cannot decide which path matters; that responsibility remains human. There is no user manual because AI is not a product with fixed behavior, but a responsive capability. Reliability now depends on humans supplying what manuals once did: intent, constraints, and success criteria.
AI does not remove thinking from the process – it reveals whether it was ever present.
Frank Ng is a retired NASDAQ CEO, who co-authors this column with his son Ryan after publishing their book Hey AI, Let’s Talk!
