A year ago, nobody asked who was paying for the AI in a design project. It was a bit of autocomplete here, an image generation there, and it disappeared into the software subscription. That's no longer true. AI features in design tools now run on credits, credits run out, and the bill depends on which workspace the work happens in.
For a product team, that's a budgeting question. For an agency, it's a business model question. We work across many client files and workspaces at once, and every project raises the same thing: whose credits is this running on, and who pays when they run out? Here's how I think about it at Breem.
Key takeaways
- AI usage in design tools is now metered, and agencies feel it first because they work across many workspaces.
- Decide at kickoff whether a project runs in the agency's workspace or the client's. Both have real trade-offs.
- Pick one of three billing models: absorb the cost, pass it through, or bundle it into a retainer.
- Add contract language about AI tool use, client data and ownership of generated assets.
- Track usage per client at the project level, not per designer.
How AI credits work now, and why agencies feel it first
Figma's help center article on managing AI credits lays out the basics: seats come with a monthly allowance, AI features draw it down, and admins can manage how credits are used. Recent updates summarized by SpyingBee go further. Enterprise admins can now set or block AI credit usage per user, and Figma Make added a model selector, so different requests can cost different amounts. Banani's breakdown also notes that monthly allowances don't roll over.
Put those together and AI usage becomes a line item: limited, variable, and controlled by whoever administers the workspace. Product teams have one workspace and one budget owner. Agencies have their own workspace plus a guest seat in several client workspaces, each with its own admin, limits and rules. A designer can run out of credits in one client's file while plenty sit unused in another.
Agency workspace vs. client workspace
The first decision on any project is where the work lives. I make it explicitly at kickoff, because changing it halfway through is painful.
Working in the agency's workspace
- Pros: we control the tools, the credit allocation and the file structure. AI usage is predictable and we can pool credits across the team.
- Cons: we pay for the credits, and at handoff we have to transfer files cleanly. Some clients' security teams don't like their product designs sitting in an external workspace.
Working in the client's workspace
- Pros: the client owns everything from day one, handoff is simpler, and the AI usage comes out of their budget.
- Cons: their admin controls our access and our credit limits. If they've restricted AI usage per user, our designers may hit a wall mid-sprint with nobody available to raise it.
Neither is right for every project. Larger clients with established design teams usually want work in their workspace. Early-stage startups often prefer we handle the tooling and hand over files at the end.
Three billing models
Once you know where the work lives, decide how AI costs are handled. I've seen three models work.
1. Absorbed
The agency treats AI credits like any other software cost and builds them into its rates. Simple for the client, and fine when usage is modest and predictable. The risk is a project with heavy generation, like lots of Figma Make prototypes, quietly eating your margin.
2. Pass-through
The client pays for AI usage separately, either by providing seats in their workspace or by reimbursing agency costs. Transparent and fair, but it adds admin, and some clients will want to approve every top-up. That slows the team down.
3. Bundled into a retainer
For ongoing work, include an AI allowance in the monthly retainer with a clear overage rule. The client gets predictability and the agency gets protection against spikes. This is the model I lean toward for retainers.
Whatever you choose, write it in the proposal. Clients in the region generally prefer a fixed scope and a fixed number. A surprise line item for "AI credits" mid-project is the fastest way to start an awkward conversation with procurement. My post on pricing design when the first draft costs nothing covers the wider pricing picture.
Contract clauses worth adding
I'm not a lawyer, and you should have yours review anything you put in a contract. But these are the topics I now make sure our agreements address:
- AI tool use: that the agency may use AI-assisted tools in the work, which ones, and any the client wants excluded.
- Client data: what client material may be put into AI tools, and what may not, such as customer data, unreleased financials or anything covered by an NDA.
- Ownership of generated assets: that generated layouts, images and code are delivered to the client on the same terms as the rest of the work.
- Workspace and handoff: where files live during the project and how they transfer at the end, including any AI-generated prototypes.
- Costs: which billing model applies, and what happens when usage goes over.
In MENA, larger clients often have procurement and IT security reviews that now ask about AI. Having these answers ready in a standard document saves weeks.
Tracking usage per client without micromanaging
You need some visibility into AI usage, but nobody wants a designer checking a credit meter before every prompt. What works for us:
- Track at the project level. Note roughly how much AI usage each project consumed at the end of each sprint. You're looking for patterns, not policing individuals.
- Separate seats or workspaces by client where possible, so usage naturally lands in the right bucket.
- Flag heavy phases in advance. Prototyping and exploration weeks burn more credits than polish weeks. Plan for them.
- Review once a quarter. Compare AI costs with what you priced in, and adjust rates or retainer allowances.
[Waleed: add how Breem currently tracks usage, e.g. a column in the project tracker, and one lesson from a recent project.]
Closing thoughts
AI credits are a small number on most invoices today. But they're variable, they're controlled by whoever runs the workspace, and agencies sit in the awkward middle. Deciding where the work lives, picking a billing model and putting a few lines in the contract takes an hour at kickoff. It saves a lot more than that when a designer hits a limit on a Thursday afternoon before a client review. If you're setting up AI workflows across several clients, my post on the client context pack covers the other half: what the agent knows, not just what it costs.
Sources: Manage AI credits (Figma Help Center), Figma pricing and credits explained (Banani), Figma product updates, October 2026 (SpyingBee).
FAQs
Who should pay for AI credits on agency projects?
It depends on where the work lives and how you price. Agencies typically absorb the cost in their rates, pass it through to the client, or bundle an allowance into a retainer. Agree which at kickoff.
Should agency designers work in the client's Figma workspace?
Often, yes, for larger clients that want ownership from day one. Just confirm that the client's admin will give your designers enough AI credits to do the work.
What should a design contract say about AI?
Which AI tools may be used, what client data can go into them, who owns generated assets, where files live, and how AI costs are billed. Have a lawyer review the wording.