Cheaper Models, Bigger Bills
Cheaper models mean lower AI bills they say but the opposite is true. Even if today's frontier models get 10x cheaper, companies will spend more per employee than they do now, not less.
This is Jevons paradox. When something gets cheaper, usage climbs, and total spend climbs with it. Cheaper steel built skyscrapers. Cheaper compute put a datacenter behind every app. Each drop in the price of inference makes another batch of work worth tackling, and the new demand outruns the lower price.
There is no upper bound on how much a company can hand to AI. The work does not run out. It expands to fill the budget.
The only real limit is the floor, and the floor is set by your competitors. Your AI spend is not decided by what AI is worth to you on its own, but by how much your competitors get out of it. If the company next door runs agents across its whole operation and you do not, you fall behind. Their output is higher, their iterations are faster, their coverage is wider. Spending less is not a saving. It is ground lost to the rivals who kept spending.
It does not matter whether the spend makes sense for you today. Plenty of companies could reasonably decide that LLMs cannot do their real work yet, and they would be right. They still feel the pressure. The worry is not what AI can do for them now, but what it might do for a competitor first. That fear shows up the moment a serious rival starts spending, and it pulls the rest of the market in. Even a business with no real use for AI will budget for it, because being the one who waited too long is not a bet worth making.
So the question is never how to spend less on AI. It is how to get more out of what you spend. The winners will not be the ones who spent the least. They will be the ones who turned each token into the most output.
A practical note: if spend only goes up, the advantage goes to whoever extracts the most from each token. ChatBotKit is built for that - the right model per task, tight instructions, and only the tools each agent needs, so your budget buys output instead of waste.