They ask: How does AI change the economics of a software product?
You say: It breaks the assumption the whole industry was built on.
Traditional software has almost no cost per additional user. You build it once, and each new customer adds revenue while adding close to nothing in cost. That is why margins were extraordinary and why the sector could grow the way it did.
A product with a model behind it does not work that way. Every interaction consumes computation, and computation is billed. The same customer asking the same question twice costs twice. So cost now scales with usage rather than sitting flat.
That changes what the business needs from engineering. Usage is no longer free growth. It is expense, and somebody has to manage it.
The answer that ends the conversation early: "it is more expensive to run." True and shallow. The point is not the size of the bill, it is that the bill now moves with how much people use the product, which is the opposite of how the industry priced everything.
They ask: Why are these tools so cheap right now?
You say: Because you are not paying what they cost, and that is a strategy rather than a discount.
The same thing happened when companies moved to rented infrastructure. Early pricing was extremely attractive. Everybody moved, built everything on top of it, and made it structurally difficult to leave. Then prices went where prices go.
So when I evaluate one of these tools, I am not only asking what it costs today. I am asking how hard it would be to move off it in two years, and whether I am building something that can be lifted.
The answer that ends the conversation early: budgeting from current pricing as though it were stable. Anyone who has watched an infrastructure bill over five years will hear that immediately.
They ask: Is your company seeing value from AI?
You say: Some, in specific places, and I would be careful with the general claim.
The figure I use in class comes from one research group looking at corporate deployments, and the share showing no measurable return was very high. I quote it as their number rather than as established fact, because that is what it is.
But it matches what I see. Companies buy access, hand it out, and measure nothing. Activity goes up. People report feeling faster. Whether anything shipped sooner or cost less is usually not being tracked.
The teams getting value are the ones who picked a specific expensive process and measured it before and after.
The answer that ends the conversation early: either repeating a dramatic statistic as fact, or dismissing the whole thing as hype. One makes you look credulous and the other makes you look like you have not used the tools.
They ask: How would you measure whether the team is using AI well?
You say: Not by how much they consume, which is the metric that spread first and is now being walked back.
Some organisations started treating usage as a proxy for productivity. If the tools make you faster, the reasoning went, then heavy use means you are moving. So consumption got measured, and people did what people do when a number is watched. They found ways to make the number bigger.
It is like judging a kitchen by how much gas it burns. You will get a very hot kitchen. You will not necessarily get dinner.
What I would measure is the work. Did the change ship, did it hold up, did the defect rate move.
The answer that ends the conversation early: proposing usage dashboards as your monitoring plan. You have described the metric the industry is currently regretting.