How Do I Keep an AI Agent's Credit Use Down?

If you hold the budget for a team's AI agents, or you're watching your own monthly plan, the number to watch is credits. On October 7, 2026 I told the group that the platform I teach on uses credits instead of tokens, and that it calculates credits much the way the large language labs calculate tokens (47:34). Credits are what you spend. Below are six things I do or showed to keep that number down. For what a credit is, see What Are Gumloop Credits?.

1. Do you have your goals before you build?

Goals come first. On October 7 I told the group that after the persona, the next thing is to tell the AI agent what its goals are:

From the recording, 9:51. The next thing that we want to do is we want to tell the agent what its goals are.

In the first video of my setup series I said what skipping this costs: without goals, setting up an AI agent or a team of them is really confusing, and you waste a lot of tokens and time.

2. Which model do I pick?

I typically build my new AI agents on the Kimi K3 model, and I gave the reason on October 7:

From the recording, 22:42. It tends to give me great results and it doesn't use a lot of tokens inside of Gumloop.

From the recording, 22:54. I find that I get more bang for the buck when it comes to using the Kimi K3 model.

That's the model I used that day. Models change, so use it as an example of the habit and not as a rule. For how to choose, see Which Model Should You Pick for an AI Agent?.

3. What do you do when a model gets too expensive?

Move it. At the top of the session I said you can move the model for your AI agent from one large language model to the next in one click:

From the recording, 7:25. And if one model becomes too expensive, you can actually slide over to the next model easily.

Later I pointed to where that happens: the model dropdown on the AI agent's screen (22:00).

4. Does the AI agent need to run every day?

Only on the days it has work. For my meeting prep AI agent I set a schedule from Sunday through Thursday in the evening, because I typically don't take meetings on Saturday and Sunday. In part one of my setup video I said I did that to be responsible with my token usage, and that on Friday night and Saturday night the AI agent isn't running and isn't using any credits. To set a schedule, see How Do I Schedule an AI Agent to Run on Its Own?.

5. Is the ask-questions ability on?

Keep it on. In part two of my setup video I said to keep it on, and why: it lets the AI agent clarify with you, so it isn't burning or wasting tokens. I showed all the abilities in that video; see Which Abilities Should I Turn On for My AI Agent?.

6. Where do you read the credits it used?

At the top of the screen, next to the dashboard name. I showed it on October 7:

From the recording, 47:21. Next to it is the G icon and the number of credits I've used setting up this agent.

From the recording, 47:27. Now 681 credits is not a lot for setting up agents.

That was the count on my screen on October 7, 2026 for setting up one AI agent. I also said the trial account gave the whole group 20,000 credits to use (22:48).

Gumloop's documentation says the running credit total sits next to a task's name in its header, and that selecting it opens a drawer with a row for each credit type charged. Its troubleshooting list for high credit costs says to switch to a cheaper model for simple tasks, start new tasks instead of long threads, reduce the number of tools, and write clearer prompts. Checked on 2026-11-05 against Gumloop's Agents documentation. Those three tips about long threads, tools and prompts come from Gumloop, not from my session.

On my relationship to Gumloop, I said this:

From the recording, 33:22. I don't work for Gumloop. I found this tool over the summer through Lenny's podcast, and I started using it and have since made myself an ambassador for them. I've created a relationship with Gumloop. I am not an employee.

More on that answer: Do you work for Gumloop? (the answer from the session).

What goes wrong

Two things, both from my setup videos.

No goals. Without them you waste a lot of tokens and time.

An AI agent that runs when there's nothing for it to do. That's the reason I don't schedule mine for the weekend.

The recording doesn't show the credit drawer opened, a price, or what any plan costs, so the only figures here are the two I said on October 7: 20,000 trial credits and 681 for one setup.

Where this comes from

This article grows from the session I taught live with Women Applying AI on October 7, 2026. The quoted passages are my own words, taken verbatim from the recording. The steps, tables and checklists around them were drafted by my team's AI agents from those words and my published pages, and checked against the recording before publishing. The lines about goals, the schedule and the ask-questions ability come from my published setup videos on YouTube and are restated, not quoted. Platform facts that have a public source carry the dated line above, and screens change.

Build Your First AI Agent in One Hour: the full session with the clips, the prompts and the transcript.

Related reading: What Are Gumloop Credits?, Which Model Should You Pick for an AI Agent?, How Much Should You Budget to Learn AI? The Gas Money Rule.