On March 30, 2026, a Figma user opened a thread on the company's forum. Their admin dashboard showed AI credit usage of 6,266 out of a 3,000 monthly limit. Next to it, a tooltip: "Full seat limits weren't enforced yet."
Their account of what happened: the balance hit zero in March, and Figma Make simply kept working. No block, no cap. Then enforcement started, and they discovered how far past the limit they'd gone.
Figma's reply on the thread was clear — and, on the facts, fair. Usage "was tracked and visible in the product, even though limits were not yet fully enforced." Nothing above the limit before March 18 was billed. And the overage would "not carry over as a negative balance into the next period." Nobody lost money.
So why start here? Because this is what a well-run rollout looks like. And it still produced a customer at 209% of their limit, confused enough to post about it. If it happens at Figma, it will happen to you.
Figma did almost everything right
Look at the sequence.
- December 9, 2025 — Figma announces per-seat AI credits and sets enforcement for March 18, 2026. More than three months of notice. Today, a Full seat gets 3,000 credits a month on Professional, 3,500 on Organization, 4,250 on Enterprise.
- The same day — individual users can see their own balance and reset date. Admins get a dedicated AI credits page.
- March 11, 2026 — a week before enforcement, admins get ways to buy more: a monthly credit package, or pay-as-you-go up to a spending limit (Organization and Enterprise first; Professional followed in May).
- March 18, 2026 — limits go live. Run out, and paid AI features stop "until credits reset or an admin adds additional credits."
Announce, show, measure, then charge. That's the textbook order.
And the business result held. In its Q1 2026 prepared remarks, Figma's CFO said over 75% of Org and Enterprise users who had been over their credit limit kept using credits, and over 95% of them stayed active. On the Q2 call, he said over 80% of paying customers worth more than $10,000 a year were consuming AI credits every week.
The rollout worked. That's exactly what makes the friction worth studying.
The complaints weren't about price
Read the forum from March onward and a pattern shows up. Very few people say "this is too expensive." They say "I can't tell what anything costs."
- March 26 — an account dropped from 3,000 credits to 1,220 in about ten prompts. One prompt cost 136.
- June 27 — after six Make pages used a full 3,000, a single prompt used 1,000 purchased credits. The user's line: "Without transparency, it is impossible to estimate or manage AI usage effectively."
- March 10 — a feedback thread titled "Figma Make AI credit limits not feasible" collected 188 replies.
Figma gave people a balance. What a balance can't show you is the price of the next action. When one prompt costs 136 credits and another costs 1,000, the balance is a rear-view mirror.
What a warm-up period is actually for
Most teams treat the warm-up as a grace period: we measure, you don't pay yet. That's half of it.
The other half: it's the only window where your customer can learn the exchange rate between their work and your meter — for free. Every warning they see, every "this would have used 140 credits," every moment they hit zero and notice it — that's the education. Skip it, and enforcement day becomes the lesson.
Figma's case shows the trap hiding inside a well-meant warm-up. When a product keeps working past zero without saying so loudly, it teaches the user something too: the limit isn't real. Then, on the day it becomes real, the same behavior gets blocked.
A warm-up period isn't free usage. It's the time your customer has to learn what their work costs — and you only get it once.
Where the order matters
The playbook I built for this lays out six steps: define the metric, align the executives, prepare the customers, build the infrastructure, let the data set the price, launch small and iterate. Figma's rollout points at three places where getting it wrong costs you.
Step 1: The metric has to be predictable in the customer's words
Credits are your unit, not theirs. If the same kind of action can cost 136 or 1,000, nobody can budget. Show cost in the customer's language — "about 40 prototype iterations left this month" — and show the price of an action before it runs, not after.
Step 3: The warm-up has to behave like the real thing
During the warm-up, let people go past zero — but tell them every single time. "You'd have been paused here. You're at 209% of your monthly limit." Same visibility as enforcement, no penalty. The customer rehearses the real meter instead of meeting it cold.
Step 6: Controls ship before enforcement, not after
Figma gave admins purchase options a week before limits went live. On the Q2 call on August 5, the CFO said: "Just this week, we began rolling out user level limits, giving admins more precise control over how AI credits are allocated across their organizations." The right feature — arriving almost five months after enforcement. Spending controls are what let an admin say yes to a meter. Ship them first.
Figma can afford a rough week. Most products can't.
Figma's Q1 remarks described the goal this way: "for our monetization model to support adoption rather than constrain it." By its own numbers, it's working. But Figma had brand trust, a product people open every day, and a market position to lean on while customers learned the meter.
A younger product launching credits doesn't get that cushion. The customer at 209% of their limit doesn't post on your forum. They just leave.
Your next pricing launch might not fail on the price. It might fail on the weeks before it.
So, before your meter goes live: could your customers predict next month's bill from what they did last month?
Free companion
Launch It — The 6-Step Framework
The six-step execution framework in a single PDF — choosing a metric that passes the four tests, the stakeholder fear map, the warm-up script, the minimum billing stack, the signals to watch after launch, and a one-page launch checklist.
Get the framework →This is one piece of a longer framework I teach in Chapter 5 of Product Strategy in the AI Era — including the hybrid model, how much volume to include in every tier, and how to remove the "taxi meter" feeling so users explore before they pay.
Sources
- Figma Forum, "Understanding AI credits" (December 9, 2025): forum.figma.com
- Figma Blog, "Updates to AI credits in Figma": figma.com/blog
- Figma Help Center, "How Figma AI credits work": help.figma.com
- Figma Forum, enforcement update (March 11, 2026): forum.figma.com
- Figma Forum, 2x overage report (March 30, 2026): forum.figma.com
- Figma Forum, "Urgent: unexpected AI credit usage issue" (March 26, 2026): forum.figma.com
- Figma Forum, "Why is Figma Make AI consuming credits so quickly?" (June 27, 2026): forum.figma.com
- Figma Forum, "Figma Make AI credit limits not feasible" (March 10, 2026): forum.figma.com
- Figma Q1 2026 Prepared Remarks: q4cdn.com
- Figma Q2 2026 earnings call transcript (The Motley Fool): fool.com