Elon Musk says his AI agent costs too much for what it delivers. His own company showed at the same time how the same agent saved $14,220. Both are true, and that is the lesson: what an agent is worth depends on the work you give it, not on what the subscription costs.
What happened
A customer of Grok Bot, the xAI agent, had used up his credit with four days left until the reset. Asked to top it up early, Musk replied that the cost of offering Grok Bot was "unsustainably high" relative to revenue.
At the same time, xAI reported in a procurement analysis from September that Grok Bot had found 43 paid licences for one software package that had gone unused for 90 days. After checking and scaling down, that saved $14,220.
Nate B Jones turns this into a framework in his newsletter and an accompanying video. His point: the cost of the work, the price you pay for it and the value of the result are three different numbers. An agent can deliver a lot for the user and still be too expensive for the provider at the current price.
In the video he adds three questions to judge the outcome. Did the agent find something you would otherwise have missed? Did it finish something that would otherwise have stayed on the pile? And how much time did it cost you to check the result and correct it?
Both pieces are by the same author and rest on a single statement from Musk. The xAI figures are the supplier's figures about itself.
What this changes for you
For a small or mid-sized business, this is not news about Musk but about your own budget. When AI providers change their prices and limits, and they do, you need to be able to say which work is worth the money.
Ask the three questions per task. Did the agent find something you missed, did it finish something, and how much checking time did it cost? An agent that rewrites a report nobody reads is expensive, however low the price.
Put a value on your own hour. Jones works with a range of $20 to $50 per hour, but pick a number of your own. Then you know when an agent pays off.
Count your attention. A pricier option that needs less checking can be cheaper overall than the cheapest one. Jones shows this with a weekly report, where a $20 option beats a $5 one once your own time counts.
Give a new idea a budget up front. Decide how much money and how much human attention you will spend to find out whether something works. Then you also know when to stop.
This is how we look at AI work too: per task and per outcome, not per licence.
How it works under the hood
You pay for an agent per token, the unit in which a model reads and writes text. Every step an agent takes costs tokens: reading the context, calling a tool, writing an interim result, reasoning again. Long tasks with many steps and a lot of context add up fast, even when the price per token is low.
So whoever deploys agents measures cost per completed task instead of per call. Set a maximum for the number of steps and the budget per run, let a cheaper model handle the simple steps, and track how much of the output a person had to correct. Without that last measure, "task completed" says little.
This news touches chapters 7, 24 and 25 of the course AI(l) or nothing by Newway. If you follow the course, this subject returns there at a calmer pace.
What you can do with this
Want to know what this means for your business? Plan an introduction: half an hour, no sales pitch, but an honest answer to the question where to begin. Rather read on first? The course AI(l) or nothing is free and starts with the basics.
Written in response to publications by Nate's Newsletter and Nate B Jones on YouTube.
