Figma promotional image showing several AI agent tasks running at once on a mobile app design — Figma's agent can now run alongside teammates' agents on the same canvas. Image: Figma via The Verge.

The Figma AI agent is no longer a beta feature. Figma moved its design agent to general availability on October 6, 2026. It also started charging AI credits for it on the same day.

The Verge's Jay Peters reported that Figma's AI agent for product design is now generally available. The agent first appeared in beta in May. This release adds library guidelines, file search inside chat and a way to watch what your colleagues' agents are doing.

I coordinate projects where designers and developers share the same Figma files every day. So I care less about the demo and more about the practical stuff. Who gets it, what it costs and where it can go wrong. That is what this guide covers.

What Changed With the Figma AI Agent on October 6

Figma's release notes say the agent is exiting beta with lower latency and a handful of new features. Here is the short list.

Library guidelines. You can upload Markdown, plain-text or JSON files to a published design library. The agent reads them every time someone prompts with that library enabled.
Live agent views. You can watch teammates' agents work on the canvas. Click the chat pile to see or follow them.
File search in chat. The agent can find files across Figma. It can insert or reference design files, FigJam sticky notes and Figma Slides content.
More ways to add context. Use the + button in the prompt box to attach Figma files, or paste Figma nodes straight into the prompt.
Generative plugins. These are now generally available, so you can pick an image model and edit images through the agent.
Custom skills. The agent can help you brainstorm, create and publish skills. Skills are now available on the Starter plan too.

Guidelines are the feature I would look at first. Figma's help center says the combined size of all guideline files in a library cannot exceed 200 KB. You need edit access to the library file, and you add them from the file menu under Library, then Manage guidelines.

Why guidelines matter more than they sound

Most agency design systems live partly in people's heads. Rules like "never use the brand orange for errors" rarely sit in the component itself. Guidelines give the agent that missing context without anyone retyping it in every prompt.

For agencies, this also means one guideline file per client library. That keeps Client A's rules away from Client B's work.

Which Plans and Seats Get Access

Access depends on your plan and seat type. Figma lists it like this:

Plan or seat	Agent access
Full seat on Professional, Organization or Enterprise	Yes
Some Education plans	Yes
Starter plan	Limited
Collab, Dev and View seats	In their drafts only

That last row is easy to miss. A developer on a Dev seat can try the agent in drafts but cannot run it on the shared client file. Plan your seat mix before you promise anything to a client.

How Figma AI Agent Credits Work Now

This is the part that will hit budgets. During the beta, the agent did not use any AI credits. From October 6, it does. Figma explains the rules in its AI credit updates FAQ.

One shared budget. The agent draws from the same credit pool as Figma Make and other generally available AI tools.
Usage varies by task. Credit use depends on prompt complexity, the actions the agent takes and the context you add, like connected libraries or attached files.
You can hit zero. If someone burned through credits in Figma Make, the agent may show 0 credits left. Limits reset monthly.
Admins can buy more. Teams can add credits through a subscription or pay-as-you-go billing.

There is some good news. Since August 25, 2026, credit add-ons include 2x more credits on Professional and 1.6x more on Organization and Enterprise, at the same price. Seat credits did not change.

If you are on Organization or Enterprise, download your beta usage CSV now. You will find it under the Admin tab, in the AI credits section. It shows one row per user per metering period, so you can see who leaned on the agent most during the beta. That is your best guide for estimating real monthly costs.

How to bill clients for agent time

Credits are a new line cost, much like stock photos or paid plugins. Decide early whether you absorb them or pass them through. My advice is to track credits per project for the first month before you change any quotes.

Figma Design file showing updated screens for adding a guest host, with member list rows and access settings — Figma's sample workflow shows screens the agent helped bring up to date with the current design system. Image: Figma.

Where the Figma AI Agent Helps an Agency Team

Figma published two detailed workflow posts alongside the launch. They show the agent doing maintenance and review work more than creating screens from nothing. That matches what I see in real projects, where the boring work eats the hours.

1. Design system cleanup

Figma's example uses a /repair-library-usage skill. The agent compares selected frames against the current library and swaps deprecated components where a clear match exists. When the right choice is not obvious, it leaves the frame alone and adds a note explaining why it stopped.

It can also rename messy layers. Think of a library where the same state is called hover, hovered and mouse-over. The agent can find those and apply one naming rule across the file.

2. Documentation for developer handoff

Figma says Uber product designer Ian Guisard built skills to document components across Uber's seven platforms. His /create-api skill writes a developer-friendly description of how a component should work. A /create-color skill maps which design token each part uses.

For agencies, this is where the real time savings sit. Clear specs mean fewer "what did you mean here" messages from developers.

3. Sorting design review feedback

After a design review, comment threads pile up. Figma shows the agent grouping repeated feedback, flagging contradictions and listing open questions. The designer still makes every product call. The agent just turns a long pile of comments into a short brief.

4. Bringing old screens back on-system

At Granola, product designer Paavan Buddhdev imports live app screens using the Figma Chrome extension. The imported buttons lose their link to the design system. He selects one and asks the agent to swap in the matching library button.

This is useful for redesign projects. Many clients come to us with a live product and no clean Figma source.

Risks and Limits to Plan For

The agent is useful, but it is not a designer. Here is what I would watch.

Silent changes. Ask the agent to propose changes before it makes them. Figma's own example prompt ends with "Don't make changes."
Client data. Check your contract before you attach client files, briefs or compliance documents as context.
Credit surprises. One heavy user can drain the shared pool for the whole team.
Accessibility gaps. Generated screens still need checks for contrast, focus states and touch targets against the W3C's WCAG 2.2 guidelines.
Weak guidelines. Vague rules give vague results. Write guidelines like you are briefing a new junior designer.

We compared earlier AI design helpers in our Figma AI guide comparing Claude and Gemini. The pattern holds. Output quality tracks the context you feed in.

Abstract illustration of a bright star shape breaking through rows of orange and black lines — Figma pitches the agent as a way to take design system upkeep off designers' plates. Illustration by Cynthia Alfonso for Figma.

A Two-Week Pilot Plan for the Figma AI Agent

Do not roll the agent out to every client file on day one. Here is a simple plan I would use.

Pick one internal library. Use your own design system or a demo project, not live client work.
Write a guideline file. Cover naming rules, spacing tokens, color use and two or three anti-patterns. Keep it short and specific.
Choose three tasks. For example, a token audit, a naming cleanup and component documentation.
Run each task twice. Once with the agent and once by hand. Record time spent and errors found.
Track credits daily. Check the AI Balance menu so you know the real cost per task.
Review with developers. Ask your front-end team if the generated specs are actually usable.
Decide and document. Write down which tasks the agent may do and which stay manual.

If your team already uses AI inside documents and slides, the same pilot habits apply. We covered a similar rollout in our Claude for Google Workspace guide for agencies.

What the Figma AI Agent Means for Agencies in India and the USA

Agencies in India often run design and development for clients in the USA across time zones. Handoffs happen overnight. Live agent views and file search can help the next shift see what was done and why.

US clients will start asking about it too. Some will want faster design system work at a lower cost. Others will ask hard questions about where their files go. Have an answer ready for both.

The bigger shift is in roles. Designers will spend less time on repairs and more time on judgment calls. Junior designers still need to learn the basics, though. An agent can swap a component, but someone has to know why it was wrong.

Pricing may change as well. If the agent cuts documentation time, clients will expect that saving to show up somewhere. Be open about where AI helps and where human review still takes time.

Key Takeaways

The Figma AI agent became generally available on October 6, 2026, after launching in beta in May.
New features include library guidelines, file search in chat, live agent views and custom skills on Starter.
It now uses AI credits from the same pool as Figma Make. Beta usage was free.
Full seats on paid plans get full access. Collab, Dev and View seats can only use it in drafts.
Start small. Pilot it on an internal library, write clear guidelines and track credit use per task.

Planning to bring AI design tools into your client workflow? Our team offers free software development consulting to help you map the process, costs and risks.

About the Author

Nitin Kumar is the Lead Project Coordinator at Debugged Pro. He works with design, development and QA teams across India and the USA to keep software projects on track. He writes about the tech news that affects agencies and their clients.