AI keeps getting cheaper. Most AI bills keep going up.
I build Elis Token Saver, which lowers what AI work costs. Set it up in a few minutes: https://www.elistokensaver.com/tutorials/dev/token-saver-mcp/?utm_source=other&utm_medium=paid&utm_campaign=crosspost-p-099&utm_content=ad-2fo7vd
Both of those are true at once. This started as a short LinkedIn note in AI bill questions, my weekly series for people who fund or buy AI. An AI agent expanded it into this longer version.
The price of one answer keeps falling
Ask a given AI the same question today and it costs less than it did a year ago.
Providers keep cutting prices. The smaller, cheaper versions keep getting better at work that used to need the biggest one.
If bills followed the price list, they would be shrinking.
The bill follows the work
They don't shrink, for two reasons.
First, every price cut invites more work. When each answer is cheap, teams stop rationing it.
The AI summarizes every support ticket, not just the hard ones. An agent, meaning an AI that works through a task on its own, gets a second and third pass to be safe. A feature that asked the AI once a day now asks it on every page load.
Second, many teams switch to the newest, priciest version the week it ships. It is better, so it becomes the default for everything, including the easy work the older version handled fine.
So the price list goes down while the monthly total goes up.
Economists have a name for this. When something gets cheaper to use, people tend to use much more of it: the Jevons paradox, first described about coal in the 1800s.
Three questions that say more than the price list
Whether you run an AI product or pay for AI on your team, these three questions tell you more about the bill than any price announcement:
- What does one finished piece of work cost today, compared with six months ago? Count the finished job, such as a ticket closed or a report written, rather than a single answer.
- How much of the spend goes to simple work a cheaper option could do? Think of small, repeated jobs, like reformatting text or a quick yes-or-no check.
- As usage grows, does the cost per finished job fall, or only the price list?
What the answers tell you
A team that can answer the first question with real numbers knows where its money goes. It can tell when a new version is worth paying for and when it isn't.
A team that can only point to falling AI prices is counting on the market to fix its margin.
The market sets the price of one answer. It does not decide how many answers you ask for, or which version gives them. That part is a choice, and it is yours.
About me
I'm Noah Salas, founder of Elis AI in Houston. Elis Token Saver connects to the AI tools teams already use - Claude, ChatGPT and Codex - and splits each job into steps, so the expensive AI only does the steps that need it. Your tools, providers and models stay as they are.
What does one finished job cost your team today, and do you know what it cost six months ago?
Book a call: https://calendly.com/thenoahsalas/30min
