Figma's Agent Credit Meter | Pixel Audit
Show notes
What the episode covers
Figma's credit screen for the design agent and Weave tools, which leave open beta today, is a meter running before you have said where you are going, and Maya argues it deserves the same suspicion as a sketchy client paywall. In this Tuesday episode for week 41 of 2026, Pixel Audit hosts Grant and Maya run a UX and UI design critique of the shared AI credit pool that now covers the agent, Weave and Figma Make.
This product design audit sticks to confirmed facts and says so when something is unpublished. The design elements discussed:
- The pricing page pool per seat, a known wallet size with no per-prompt cost
- The conflict between release notes and the help article on Starter eligibility
- The left-rail and on-canvas entry points and the @-mention flow for tokens and components
- Collaborator-visible chats and how a shared screen can shape behavior
You will walk away knowing how to run the client-paywall test (cost before action, clear balance, escape hatch) on any metered tool, and how to tell a protective pause from a taxing one.
Timeline
In this episode
7 moments worth skipping to. The timecodes match the player above.
- 0:12Introduction
- 1:31What Flips at GA: Confirmed Facts
- 4:54Entry Points and the @-Mention Flow
- 8:42Quick-Fire: Red Flag or Green Flag
- 11:04The Turn: Prompting With a Meter Running
- 14:45Honest Meter or Quiet Paywall?
- 17:32Outro
Quick answers
Straight from the episode
The questions this one settles, without the listen.
- When does Figma's agent move from free beta to credit-based pricing?
- The agent beta began May 20, 2026 and was credit-free. Credit limits have been enforced on Full seats since March 18, and at general availability the agent's use flips to credits. The hosts treat the free beta followed by credits as a green flag, with a yellow on timing.
- Who gets access to Figma's agent, and is Starter eligible?
- Sources conflict. Figma's release notes and a help article disagree on Starter eligibility, which the hosts flagged red. Collab and Dev seats are described as drafts-only, Professional has pay-as-you-go, and beta users keep their access.
- How do credits work for the Figma agent, and what does each prompt cost?
- The pricing page lists a baseline credit pool per seat for Starter, Professional, Organization and Dev/Collab. The hosts read it as a shared allowance, so the wallet size is known. The per-action or per-prompt cost is not published, and that gap is the main concern.
- Where does the agent live in Figma and how does the @-mention flow work?
- Figma's launch post describes the agent as embedded in Figma Design, working with team design systems and handling feedback. It is reached through left-rail and on-canvas entry points, and you @-mention tokens and components. Chats are visible to collaborators, and no pre-prompt cost screen is known.
- How might credit meters change the way designers prompt?
- The hosts argue hesitation is a hidden cost of iteration. A shared pool could make juniors cautious or lead teams to hoard credits, though they frame this as speculation, not data. The Reddit anecdotes cited are non-agent, pre-GA reports.
- Is Figma's agent credit system an honest meter or a quiet paywall?
- The verdict is split. Maya wants the price shown on the button before an action. Grant accepts a clear balance and a cheap refill, including the June 3 pay-as-you-go entry. The hosts tested it against a client-paywall standard: cost before action, clear balance, and an escape hatch.
Transcript
The full conversation
Every word of the episode, 3,163 of them, in the order they were said.
Read the transcriptHide the transcript
GrantPicture getting into a cab where the meter's already ticking before you said where you're going.
MayaUgh, I hate that cab.
GrantNow picture it's your design tool.
MayaOkay, that visual. Wait, which tool?
GrantFigma. Today, October 6th, the design agent and the Weave tools leave open beta, and they start drawing from the same AI credit pool as Figma Make.
MayaSo every prompt costs something? Um, every single nudge of a frame?
GrantThat's the question. Yeah. This is Pixel Audit. I'm Grant, and that's Maya.
MayaAnd I'm attacking today. I'm treating that credit screen exactly like a client's paywall, sketchy until proven otherwise.
GrantWhich makes me the defender. I have to find the reasoning behind the Meter.
MayaGood luck. I mean, come on, a tool that makes you hesitate before you iterate?
GrantOr honest pricing for something that doesn't cost the same every time. I honestly don't know yet.
MayaOoh, Unresolved, my favorite.
GrantAnd that's the deal today. No hot takes we can't back up, and no inventing screens we haven't seen. If we can't state it, we say so.
MayaHonest Audit. I can live with that, as long as I still get my dramatic pauses.
GrantBefore we judge the Meter, let's pin down what's actually confirmed about the GA switch.
MayaFine. Facts first, then I get to be mean.
Speaker 3Okay, so the Meter's running, but what does it actually say on the dial? I keep hearing credits, and I have no idea what a credit buys.
GrantFair, and I'll tell you right now, the agent's own cost per action isn't something I can give you. What I can give you is the pool it draws from and the timeline.
Speaker 3Timeline first. When did this thing even show up?
GrantFigma's release notes say the Agent entered beta on May 26th, 2026. Same notes say AI credits apply at general availability. So beta was the free sample.
Speaker 3Free free or free-ish?
GrantFree. Figma's launch post for the Agent confirms it consumed no credits during beta, and a help center article says the same thing, no credits for any prompt or action.
Speaker 3Huh. So everybody who's been poking at it has been riding without a meter.
GrantMm-hmm. One detail from that help article though, credit limits have been enforced for full seats since March 18th, so the pool already existed for other AI features. The agent just wasn't charging against it yet.
Speaker 3Wait, really? So the cab's been on the street the whole time?
GrantYes, you and your cab. Yes. Now, the pool itself, Figma's pricing page lists it per seat, and the baseline is where the agent will draw from.
Speaker 3Okay, hit me. Numbers.
GrantStarter gets credits a day, capped at 500 a month.
Speaker 3That's tiny.
GrantIt depends what an action costs, and that's the unpublished part. Next, Professional full seats get 3,000 credits, and that's the $16 seat.
Speaker 3And the pricier one?
GrantOrganization full seats get 3,500 credits for $55.
Speaker 3Hold on, 39 more dollars for 500 more credits?
GrantWell, the seat buys other stuff too, Admin, governance, that side. But yeah, on credits alone, it's a funny ladder. And Dev and Collab seats get 500.
Speaker 3Same as Starter's monthly cap, basically.
GrantRight, and pricing as of the day we collected it, so check the page before you quote me.
Speaker 3So what does a pool actually mean day to day? It's a shared allowance, not a price per click, right?
GrantThat's how I read it. You get a number of credits per seat, and the Agent's work draws from that number. What I can't tell you is how many actions that adds up to because the cost per action is the missing piece.
Speaker 3So we know the size of the wallet, not the price of anything in it.
GrantExactly. Keep that in your pocket because it comes back all episode.
MayaOkay, next question, and this is the one I actually care about. Who gets in the door?
GrantAh, this is where the sources disagree.
MayaOf course they do.
GrantThe release notes say full seats on professional, organization, and enterprise get the agent. Collab and dev seats can use it in drafts only. Starter, education, and government are excluded.
MayaExcluded, flat out.
GrantExcluded, flat out, but the older help article says starter is eligible.
MayaSo Figma's own pages are fighting each other?
GrantCould be the rules changed, and one page didn't get the memo. I can't tell you which is current. If you're on starter, check before you plan a workflow around it.
MayaAlso, the notes mention pay-as-you-go credits on professional from a June three entry. So there's a way to top up.
GrantYep, and beta users keep access at GA per that same help article.
MayaOkay, facts on the table. Now let's open the product and see where the agent actually lives. Okay, so I'm opening the file. First thing I wanna know is where the agent actually lives because if it's a button I can bump with my elbow, that's a problem.
GrantFigma's launch post says it's embedded in FigmaDesign, not a separate app, not a tab, inside the canvas you already work in.
MayaRight, and it says it works with the team's design systems and handles feedback.
GrantWhich is the part I like. Feedback is where designers burn hours.
MayaSure, but hold on.
GrantGo.
MayaOkay, embedded means casual. Casual means I click it like I click anything else, and this click spends money.
GrantSpends credits.
MayaSame thing, Grant. Come on.
GrantFine, spends credits, but think about what the alternative is. You leave the file, open some pricing gated portal, paste context back in. That's friction, and friction kills the work.
MayaMm.
GrantEmbedded keeps you in flow. I spent 15 years watching people make worse decisions the second you made them switch screens.
MayaOkay, but flow is exactly the trap. A slot machine keeps you in flow too.
GrantOof.
MayaI mean it. The brief for this episode lists two doors, the left rail and on the canvas itself. Two doors, both easy to walk through.
GrantAnd you want a speed bump.
MayaI want to know if there is one. A client's paywall gets a confirmation screen. Does the agent door?
GrantHonestly, I can't tell you from what we have. The sources talk about where it lives, not what it shows before you fire a prompt. So I'm not going to invent a modal.
MayaGood. Neither am I. That's the whole Audit, right? What we can see versus what we're guessing.
GrantRight. So next door, the @mention Flow.
MayaTokens and components.
GrantYou type an at sign and you point the Agent at your own library, your variables, your button component, not some generic gray rectangle.
MayaOkay, that's smart. Wait, though. Is it?
GrantSay more.
MayaPointing at my real system means the output is more useful. More useful means I trust it. Trust means I keep prompting.
GrantWhich is how every product wants you to feel. Why is that sinister?
MayaIt's not sinister by itself. It's just, um, the better the tool, the faster the Meter feels like it runs. Good design and good billing pull the same direction here.
GrantHuh. Yeah, that's a real tension. The best feature is the one that makes you want more of the metered thing.
MayaMm-hm.
GrantAnd then there's the chats visible to collaborators.
MayaOh, I have feelings about this.
GrantOf course you do.
MayaMy prompt history in a shared file where my teammates can see it, including the sloppy ones, including the fourth try at the same thing.
GrantCould be a feature. Your team sees how you got there. Review gets easier.
MayaOr it's a security camera on my iteration habits. Designers already hide the ugly drafts.
GrantFair. I'll give you that one, but only on tone. I can't tell you whether anyone is actually watching the pile of attempts.
MayaNobody's saying they are. It's just the screen exists, so the feeling exists.
GrantThat's a fair way to frame it. The design of a shared screen changes behavior even if nobody ever opens it. Designers know that from every comment thread they've ever softened before posting.
MayaRight. And that's why it belongs in the audit. A feeling is a design outcome even when we can't measure it. I'd rather flag it than pretend it isn't there.
GrantOkay. We've been critiquing for a while, and a lot of this is what we can't see. Let's lighten it up with something fast.
MayaRed flag, green flag on the credit facts we already laid out?
GrantYes. Buzzers ready.
MayaMine's loud. Okay, flag round. I'll throw facts. You wave a color. First one, the launch post says the agent burned zero credits during beta, and credits kick in at general availability.
GrantGreen on the disclosure. Figma's launch post says it plainly. Yellow on the timing, though. Free then build is the oldest move in the book.
MayaHa. Yellow. Fine.
GrantNext.
MayaThe agent draws from the same credit pool as Make. One wallet, two tools.
GrantGreen, honestly. Separate meters per feature would be worse.
MayaWait, though. Worse for whom? If I'm deep in Make, my agent budget quietly shrinks.
GrantMm, fair. Green with an asterisk.
MayaStarter eligibility. The release notes say Starter is excluded. The older help article says otherwise.
GrantRed for the contradiction. Two official pages, two answers, and the designer finds out by trying.
MayaBy clicking the button and seeing what happens.
GrantWhich with a meter is a bad way to learn.
MayaPay-as-you-go credits on Professional. The release notes have a June entry for it.
GrantGreen. An escape hatch beats a hard wall at the end of the month. I've seen worse in car auctions where there's no hatch at all.
MayaLast fact. The pricing page lists a per seat pool. Professional full seats get three thousand credits.
GrantGreen for putting it on the page, but a pool measured in credits means nothing until you know what one prompt costs.
MayaRight. And nobody's published that, so it's a tank with no gauge.
GrantYellow then.
MayaHold on. Let me sanity check the pattern. The red was about a contradiction, not a price.
GrantYeah. Every flag we waved came from something written down. The trouble is all in what hasn't been written down yet, and a meter makes that gap expensive.
MayaOkay. The design stuff doesn't get flags, only questions. Is a left rail entry too casual for something that spends money?
GrantDoes the on-canvas door feel different? Does an @mention feel free when it isn't?
MayaAnd do collaborators seeing your chats make you prompt more carefully or less?
GrantOoh, both probably, depending on the collaborator.
MayaDepends if it's your boss.
GrantSo the facts are mostly fine. The unknown is the gauge.
MayaWhich brings it to what a credit pool does to the way you prompt.
GrantYeah, that's exactly.
MayaOkay. So I went looking for how real designers talk about iterating under this pool, not Figma's pitch. People.
GrantAnd?
MayaThe pricing explainer on Awesomics blog titled Figma Pricing in Twenty Twenty-Six: What You Really Pay quotes users from the Figma Design subreddit. One said they took about a hundred and twenty credits to build a small interactive component.
GrantThat's a button with a hover state, basically.
MayaRight. And then they spent eighty more fixing a miss.
GrantHold on. Whose numbers are those?
MayaUsers, not Figma's. And the explainer doesn't say which feature they were using, so these are not Agent costs. The Agent's per action price is still unpublished.
GrantGood, because I was about to do math with it.
MayaOf course you were.
GrantI have a spreadsheet for everything, Maya, but fine, it's an anecdote. What was the other one?
MayaSomeone said they burned through three thousand company credits in about three hours.
GrantThree hours?
MayaMm-hmm.
GrantOkay. That one I want context on. Maybe they were stress testing. Maybe it was a bad afternoon. One person's story isn't a rate card.
MayaTotally. Also, those reports predate the Agent's metering at GA, so I'm not saying that's what you'll pay. I'm saying that's how designers describe the feeling of iterating inside the pool.
GrantAnd the feeling is the interesting part, because iteration is the design process. Nobody nails it on try one.
MayaThe second try is the job.
GrantExactly. And when I was buying cars, I saw this constantly. Put a per-unit cost on something, and people change how they use it. I once watched a buyer skip a second pre-purchase inspection to save a few hundred dollars on a car worth six figures.
MayaWait, really? Why?
GrantBecause the inspection had a visible price tag and the risk didn't. The cost you can see beats the cost you can't every time.
MayaSo the eighty Credits to fix a miss-
GrantIs a visible price on being wrong, and being wrong is how you get to right.
MayaHmm. But the other side of that is a budget can make you sharper, write a better prompt, think first.
GrantSure, sometimes. I've been that guy with a spreadsheet who thinks harder because it costs money.
MayaBut?
GrantBut thinking harder before you try is not the same as thinking by trying. Designers sketch badly on purpose. That's the method.
MayaYeah. Rough drafts have to be cheap.
GrantThink about what that does to a team. Picture a newer designer watching the balance, noticing a senior one spend freely, and quietly learning which tries are allowed and which aren't.
MayaOr the opposite. Everybody hoards credits for the big idea and never tries the weird one, and the weird one is usually where the good stuff is.
GrantAnd that's the part no pricing page can tell you. It's a behavior, not a line item. All we can say is that designers describe it that way.
MayaWhich is exactly why I wanted the user reports in here, not as proof of what the Agent costs, just as how the people inside the pool talk about it.
GrantAnd here's a fact that matters for who feels this. Figma's release notes have a June third entry that adds pay-as-you-go credits on Professional.
MayaSo if you run out, you can keep going for a price.
GrantRight, which softens the wall. It also means the wall is a purchase decision, and you make it mid-flow.
MayaUgh, mid-flow is the worst time to open your wallet.
GrantIt's when you're least likely to say no.
MayaUgh. So the hidden cost isn't the credits, it's the half second where you think, "Do I really need to try that again?"
GrantThat half second multiplied by a whole team across a whole project.
MayaOkay, so is that honest pricing or a quiet paywall?
GrantFine. Time to argue it. You're up first. Sure.
MayaOkay, I'll start with what Figma says in its own defense. The Help Center FAQ on AI credit updates says that since March of twenty twenty-six, when it introduced ways to buy extra Tokens, the company kept investing in token efficiency.
GrantAnd improved its first-party models and shipped several new AI features.
MayaRight, which sounds nice, but notice what that is. It's a promise that the same prompt should cost less over time.
GrantShould. The FAQ gives me a direction. It doesn't give me a price.
MayaExactly. The taxi says it's getting more fuel efficient, and the fare is still a surprise.
GrantFair. So let's run the test I'd run on a client's Paywall. First question, is the cost visible before I act?
MayaAnd for the Agent, Nobody's shown us that.
GrantNothing I can state. I'm not going to invent a screen I haven't seen.
MayaMm. Second one, is the balance clear?
GrantThis is where I'll defend them a little. A pool of credits is a balance. It's a number. Number goes down, you can see it go down. That's more honest than a vague fair use clause.
MayaSure, if the number sits somewhere I'd actually look at it. I mean, come on, a gauge in the account settings is not a gauge.
GrantMm, okay, that I can't answer. I haven't seen where it lives on the canvas, so I can't grade it.
MayaThen we both say that out loud. Incomplete.
GrantIncomplete. Third question, escape hatch.
MayaOoh, this one matters.
GrantBecause when you run dry, what can you do? Buying more is one hatch. We've already said that exists. But a hatch that costs money is a toll booth, not an exit.
MayaCareful, that sounds like my argument.
GrantThen take it. But here's my counter. A client paywall that hides the price is a dark pattern. A meter that's simply expensive and tells you so is just a price. Those are different crimes.
MayaDifferent, sure, but what if it's both? Honest on the pricing page, murky in the moment.
GrantThen it's honest at the door and fuzzy at the register, which, uh, a lot of shops are.
MayaSee, I land harder than you. If I can't see the cost before I click, I hesitate. Hesitation is the product working against the designer.
GrantHesitation can be a good thing, though. Plenty of products add a pause before something you can't take back. The question is whether this pause is protecting you or just taxing you. I'd wanna know which one it is before I praise or condemn it.
MayaThat's the line, isn't it? A pause that protects you shows you the stakes. A pause that taxes you just makes you nervous and tells you nothing.
GrantAnd I'd say that depends on what's on screen, which we don't know. So my verdict is a shrug with a spreadsheet.
MayaOf course it is.
GrantHonest people would split here. You want the price on the button. I take a clear balance and a cheap refill.
MayaCould we get both?
GrantWe could ask for both, Loudly.
MayaFine. Left open. Now give me something to actually do with all this.
GrantYeah. One thing you can steal coming right up. Okay, the Unresolved part first. Maya wants the price printed on the button. I'll live with a clear balance and a cheap refill.
MayaAnd neither of us moved. Honest people still arguing.
GrantSo here's what you can steal. Borrow it from Paywall Review. Before any metered action, show three things: the cost, the balance left, and a way to undo.
MayaCost, balance, undo.
GrantYep. Run it on your own product tomorrow. If any one is missing, you've built a hesitation.
MayaOoh, and undo is the one people skip.
GrantAlways. Spend is easy to design. Regret is not.
MayaAnd honestly, it travels. Cost, balance, undo works on a checkout, a credit pool, anything where a click costs something.
GrantThat's the point of borrowing it from Paywall Review. Same test, different product, and it takes about a minute to run.
MayaOkay, CTA time. Screenshot a UI that bugs you and tag the show. We'll audit the good ones.
GrantSubscribe so you catch next week's audit.
MayaAnd if this changed how you look at your own designs, leave a review. It's free. I checked.
GrantUnlike some tools we know.
More episodes
Keep listening
Other episodes of UX Autopsy, newest first.
- YouTube's Subscriptions Tab | Pixel AuditNov 10, 2026 · 19 min
- Election Night Results Maps | Pixel AuditNov 3, 2026 · 19 min
- Subscription Cancel Flows | Pixel AuditOct 27, 2026 · 20 min
- Redesign Backlash Docket | Pixel AuditOct 20, 2026 · 18 min
Sources
Where this came from
6 reports behind the episode. Every one of them opens where it was published.
