Skip to main content

Credit model penalizes users for the agent's mistakes — one day of work consumed nearly my full monthly allocation

  • July 6, 2026
  • 2 replies
  • 29 views

Oleg.Lunar

I want to share feedback on the Figma Make credit system, because in its current form it is not workable for real projects.

Yesterday I spent one working day building a page in Make (Full seat, 3,000 credits/month). By the end of the day I had consumed almost the entire monthly allocation — without reaching the intended result. The core problem: the agent frequently makes mistakes, ignores constraints, and breaks previously working parts of the build. Each mistake requires another prompt to fix, and each fix consumes more credits. The more rework the agent causes, the faster credits drain. Effectively, users pay for the model's errors.

Suggestions:

1. Improve agent reliability, or stop charging full credits for correction cycles caused by the agent's own errors.
2. Add cost estimates before execution, usage alerts, and spending caps.
3. Reconsider whether 3,000 credits is a realistic monthly limit for professional iterative work — right now it can disappear in a day.

I like the product concept, but until the economics are fixed I'll be evaluating alternatives for iterative development and keeping Make only for initial generation from design context. I hope the team takes this feedback seriously — judging by this forum, I'm far from alone.

2 replies

djv
Figmate
  • Community Support
  • July 7, 2026

Hi ​@Oleg.Lunar, thanks for your honest feedback! 

A full day of credits being used for rework instead of progress is a rough experience, so I understand the frustration. Quick context: credits are charged per prompt based on complexity, model, and context/chat history in the file, not on whether the output matches expectations. When the agent produces something (even if it is not what you anticipated), that's still a charged action; model choice (Opus > default) and long chat history in a file also drive usage up. A couple of tips that can help:

  • Break requests into smaller, single-action prompts so mistakes don't compound
  • Ask the agent to describe its planned change before applying it

Check out our best practices guide for more information. Quick question: what model were you using that day?

If the usage still feels genuinely disproportionate after using these tips, please let me know. I’d be happy to help connect you with our support team to review your usage.


Adam_Ogrodnik_2

Welcome to the new world, my friend.
Incompetence has always come at a cost; now we're simply paying for the AI version of it.Â