Figma shipped three launches on October 15, 2026, all aimed at putting design, code, and content on one canvas. Specs and code layers give agents design intent and implementation. Payload in Figma adds content. This digest reads them against the question this blog keeps asking: can agents learn to design?
This digest is based on Figma’s own announcements, TechCrunch coverage, and release-note aggregators, all dated October 14–15, 2026. We did not test anything hands-on.
What shipped this week (TL;DR)
- Specs: a multiplayer Markdown container on the canvas, in open beta on paid plans across web, desktop, and mobile. Drag any
.mdfile onto the canvas to create one (Introducing Specs). - Code layers, open beta: available on all plans from October 15. Creation uses AI credits; paid plans get free creation and editing through November 4, 2026 (Design together with code layers).
- Payload in Figma: a content and data backend inside Figma Make, available on all plans, with access rolling out to all accounts over the coming weeks (Payload in Figma).
The lead story: design, code, and context on one canvas
The umbrella announcement is Figma’s “Build by design” post, credited to CEO Dylan Field (Build by design). Its framing is “for a long time, coding was expensive… AI has made coding cheap,” and it separates “building to think” from “building to ship.” Those are Figma’s words and are presented here as vendor framing.
The same batch also includes design-system guidelines (open beta), code sources (closed beta), and new MCP integrations. TechCrunch reported on October 14 that the launches came during Figma’s first Config in India, and that the Markdown editor is meant to define a company’s design rules for Figma’s own agents and third-party ones (TechCrunch). The notable part is that Figma is shipping an input format for agents, not only features for designers.
Specs: Markdown as a medium, and as machine-readable intent
Specs is a container for context on the canvas. Drag in a .md file and it becomes a spec. It can embed a FigJam file, a frame, or any canvas node such as diagrams, code layers, and designs, so the written plan and the work sit side by side. Figma’s agent can read and reply to comments in a spec. Through the Figma MCP server, agents like Codex can read a spec and see all the context it links to. The stated purpose is a shared plan that guides agents “from prototype to production” (Introducing Specs).
This is the thesis question. A Markdown spec gives an agent design intent: rules, constraints, a plan. Whether it gives the agent design judgment, the ability to tell when the plan is wrong, is the open question. A spec can say “the primary button is 44px tall,” and an agent can comply. It is much harder to write down “this layout feels cramped because the whitespace rhythm is off” in a form an agent can act on without a human in the loop. Specs is a strong machine-readable contract for intent. The launch does not answer whether intent alone is enough.
Code layers go to open beta
Code layers move into open beta on October 15 and are available on all plans. Figma announced the feature earlier this year (Code on the Figma canvas). There are several ways to create one: convert existing designs to code, bring in a coded prototype from a third-party tool, a local folder, GitHub, or the MCP server, or capture a live website with the Figma Chrome extension. Figma says the GitHub and local-folder imports cover “supported, lightweight prototyping” codebases. Design-to-code runs the other way, from a design to code.
For design engineers, the collaboration story is the useful part. Teammates can try prototypes side by side, leave feedback on them, and duplicate a code layer to explore an alternative. Figma says support for more complex codebases will come “over time,” so treat it as suited to prototypes and smaller surfaces for now.
The return path is sourced. A GitHub import stays connected to that repository. You can push updated code to a branch, and a pull request lets your team review changes before they merge. Extract designs converts a code layer’s screens or flows back into editable Figma frames (Design together with code layers). The open question is narrower than “is there an exit.” It is what happens with sources that are not GitHub, such as local uploads and live-site captures.
Payload in Figma: content and data on the canvas
Payload in Figma brings a content and data backend into Figma Make. Payload is an open-source content framework, and Figma’s post says you can take an app outside Figma once it is built (Build by design). Teams store and manage content and data, host apps, and store API keys in one place (Payload in Figma). Hosting is in beta and free for now, with pricing to be announced before it takes effect.
Specs carries intent, code layers carry implementation, and Payload carries content. The lock-in tension is real, but the sourced facts are more balanced than an “all in one vendor” story. The framework is open source and portable, while the hosted workflow lives in Figma.
What each launch gives an agent
| Launch | Input format | Where it lives | What it constrains |
|---|---|---|---|
| Specs | Markdown (.md), multiplayer |
On the canvas, alongside designs | The plan and intent an agent follows |
| Code layers (open beta) | Real code via convert, import, MCP, or live capture | On the canvas as a layer | The implementation an agent reads and modifies |
| Payload in Figma | Content and data via an open-source CMS framework | Inside Figma Make | Content that stays bound to the design system |
Read together, the three launches give an agent intent, implementation, and content. Code layers keep a GitHub import connected to its repository and can push changes to a branch, and Payload can leave Figma. What remains is how much of an agent’s working context stays inside Figma’s canvas.
Can agents learn to design? What Specs does and doesn’t solve
Specs joins a family of intent formats, alongside Google’s DESIGN.md and Anthropic’s Claude Design, both covered in last week’s digest. Each is a bet on written decisions: clear enough rules should produce good design. Taste is not something any of them has yet been shown to supply.
That bet solves intent transfer, not judgment. A spec tells an agent the rules. It does not tell the agent when a rule is producing a worse result. Taste, proportion, and the judgment to override a constraint live in the gap between the spec and the shipped screen. Treat Specs as a contract format, not a creativity source. Whether an agent that follows written rules more closely also designs better is the question for next week.
Also this week: Config India and ecosystem notes
Figma’s first Config in India is the setting for these launches (TechCrunch). The Figma newsroom lists an education program, new research, and an MCP partnership with Sarvam.
Previously (context, not news)
A recap of our October 8 digest: Figma Agent reached general availability; Google published DESIGN.md as an open spec in alpha; Anthropic shipped Claude Design; ProtoPie made its MCP integration official; and the DTCG released the 2025.10 design tokens spec, its first stable version, on October 28, 2025. Several vendors now ship their own formats for feeding design intent to agents, and none is a shared standard. DTCG remains the one to watch.
FAQ
What exactly is a Spec?
A multiplayer Markdown container on the canvas. Drag in any .md file to create one. It can embed a FigJam file, a frame, or any canvas node, and Figma’s agent can read and reply to comments in it (source).
Is Specs on all plans? No. Specs is in open beta on paid plans across web, desktop, and mobile (source).
Is code layers available on my plan? Yes, it is available on all plans from October 15. Creation uses AI credits, and paid plans get free creation and editing through November 4, 2026. The sources do not state the free-plan credit limits (source).
Can I bring existing code into Figma? Yes. Code layers can import a coded prototype from a third-party tool, a local folder, GitHub, or the MCP server, and the Chrome extension can capture a live site. Figma describes GitHub and local-folder imports as “supported, lightweight prototyping” codebases (source).
Can agents read Specs? Yes. Figma’s agent can read and reply to comments in a spec. Through the Figma MCP server, agents like Codex can read a spec and see the context it links to (source).
Does Specs replace DESIGN.md or Claude Design? Not stated in Figma’s announcements. They are separate formats from separate vendors.
Does code layers handle complex codebases? Not yet. Figma says support for more complex codebases will come “over time.” It describes the current GitHub and local-folder imports as “supported, lightweight prototyping” codebases (source).
What to try this week
Code layers open beta is on all plans, so steps 4 to 7 need no paid upgrade, though free-plan credit limits are not stated. Specs requires a paid plan.
- On a paid plan, drag any
.mdfile onto a Figma canvas to create your first Spec. - Embed a frame or design node in that spec so the written plan and the design sit side by side.
- Leave a comment in the spec and see whether Figma’s agent reads and replies.
- Convert an existing design into a code layer (open beta).
- Duplicate that code layer and change one thing: a cheap branch for exploring an alternative without touching the original.
- Import a supported prototyping repo from GitHub, make a change, and check whether the branch and pull request flow fits your review process.
- Capture a live site with the Figma Chrome extension and compare it with a design-to-code conversion.
What we don’t know
These are Figma’s announcements and press coverage. We did not test anything hands-on, so treat the behavior descriptions as vendor documentation, not evaluation. Config India content beyond the October 15 posts was not fetched. We found no DTCG news dated this week. Product Hunt launches were not verified from a primary source, so they are excluded. Open questions: how code layers handle sources that are not GitHub, the free-plan credit limits, and Payload’s pricing once hosting leaves beta.
How This Guide Was Built
This digest was produced by the Design Agent arena: a planner wrote the rubric, 13 builder models drafted independently, a judge scored each draft over three passes, and QA checked the winner against the source pages. Every link was verified live before publication. Judge scores on the shipped text averaged 7.92 out of 10, below the 8.0 threshold, so this post ships as a best-effort draft. Drafts were checked against Figma’s own pages and press coverage, and no hands-on testing was performed.