Most people use AI the normal way: they ask for what they want and hope they get it. Sometimes the agent doesn’t work hard enough and hands you something half-baked. Other times it goes the opposite direction, spending way too much time triple-checking everything and costing you money you didn’t want to spend.
Adding one sentence to your prompt fixes both:
“My budget is X. Find a way to track your token spend, and ensure you complete the task before the budget runs out.”
That’s the whole trick.
One thing to know: this only works in a real agent, like Claude Cowork, Claude Code or Codex. It won’t work in the regular Claude or ChatGPT chat window. If you’re not using an agent yet, you should be anyway.
Here’s why it works. These models charge by tokens, which is basically the amount of text they read and write. An agent doing a long task can burn through a lot of them. But an agent can keep track of its own spending as it works and plan its strategy around the budget you gave it.
And it works in both directions. Set a small budget, and the agent can plan to stay inside it. Set a big one, and it knows you want it to push further than it normally would. If you’re a company with money to spend, you can even say, “We have this much to spend, I know it’s a lot, make the most of it.”
So try it on your next big task. Hit reply and tell me how it goes.
— Matt
What should I write about next? Vote or suggest a topic →