Why Figma ships the agent layer.

The agent layer defined

Figma's agent-layer push has two distinct vectors. The first is Make, the design-automation runtime that Figma acquired in January 2024 as part of its post-IPO-collapse restructuring. Make is not a generative image tool. It is a workflow orchestrator that accepts structured design briefs — brand tokens, component constraints, layout rules — and produces compliant design variants at scale, with no designer touching the canvas for each iteration. The second vector is Dev Mode agent handoff, which Figma began expanding in Q3 2024 to allow agents running in engineering pipelines to interrogate a Figma file programmatically: retrieve token values, extract component specifications, resolve annotation conflicts, and push updates back into the file without a human intermediary.

Taken separately, each looks like an incremental feature. Taken together, they define a workflow in which the design system becomes a live data source for an autonomous software-delivery pipeline. A product manager defines a feature. A design agent generates compliant mockups against the existing design system. An engineering agent reads the mockup via Dev Mode, resolves component-to-code mappings, and opens a pull request. A review agent checks the PR output against the original design spec and flags deviations. Figma sits at the centre of every step. That is not an incremental feature. That is a new category position.

Lucia Manders, Figma's vice president of platform products, described the architecture to enterprise partners at Figma's closed Config partner session in September 2024 with characteristic directness: the design file is the source of truth for the entire product delivery chain, and the agent layer is the mechanism that lets the rest of the chain read and write to that source without queuing on human availability. Manders did not describe it as an AI product. She described it as an infrastructure decision about where authority over design decisions is stored. That framing matters for what follows.

Twelve months of buyer data

INTELAR tracked 23 Figma enterprise deployments across the Fortune 500 between January and December 2024, covering technology, financial services, consumer goods, and healthcare verticals. The deployments split cleanly along two axes: companies that adopted Make as a design-production accelerator, and companies that adopted Dev Mode agent handoff as an engineering-velocity tool. Only four of the 23 deployed both in the same quarter. That gap is the most important number in the data set.

Among Make adopters, the dominant pattern was design-at-scale: marketing and brand teams using Make to generate localised campaign assets, product packaging variants, and in-app promotional content without scaling design headcount. American Express's brand operations team, which signed a Figma enterprise contract renewal in March 2024 with Make included, reported a 74 per cent reduction in the cycle time from creative brief to approved asset — falling from 11 working days to 2.8. The team's design director, who asked not to be named in this report, described the operational logic plainly: "We were producing 340 asset variants per campaign across 18 markets. The same three senior designers were approving all of them. Make eliminated the production bottleneck without touching the approval gate." The creative director still signs off. The canvas work between brief and sign-off is now mostly automated.

Among Dev Mode agent handoff adopters, the pattern was different. These deployments were primarily engineering-led — teams using the programmatic API access to wire Figma files into CI/CD pipelines. Salesforce's platform engineering organisation, which piloted the Dev Mode agent integration in Q2 2024 across three product squads, measured a 31 per cent reduction in front-end rework caused by spec-implementation drift: the gap between what a designer specced and what an engineer implemented, which historically required a review cycle to surface and a revision cycle to fix. With the agent reading the Figma file directly and flagging deviations before the PR merged, the drift was caught earlier, at lower cost. The Salesforce team estimated the rework reduction saved approximately 180 engineering hours per quarter across the three squads. At fully loaded cost, that is a material number for a pilot that required no new tooling budget.

The competitive read

Adobe Firefly and Canva AI are the two most cited competitive comparisons in Figma's enterprise deals, and both comparisons are structurally misleading. Adobe Firefly is a generative image and vector engine built to integrate with Creative Cloud workflows — Photoshop, Illustrator, Premiere Pro. Its enterprise positioning is creative production: generating commercial-safe imagery, removing background, applying style transfers. Figma's Make is not competing for the same budget. Figma competes for the design-system-to-delivery budget, which sits in product and engineering procurement, not creative services. The two products solve adjacent problems for buyers who are almost entirely different people inside the same company.

Canva AI presents a more interesting competitive surface. Canva's enterprise product has grown significantly in adoption among marketing and brand teams — precisely the teams that Make targets for design-at-scale workflows. Canva's AI layer generates on-brand social, presentation, and print assets faster than any competing tool, and its user base is broader and less technical than Figma's. The risk for Figma is not that Canva displaces Make in design-system-connected workflows — Canva does not have access to token-level design system data — but that CMOs consolidate marketing asset production in Canva and never surface the Make use case to their design systems team. The purchasing conversation then never happens. Figma's enterprise sales organisation is actively managing this sequencing problem by targeting design system owners, not marketing operations leads, as the primary champion in Make deals. The risk is that design system owners often lack the procurement authority that marketing operations leads carry.

The more structurally serious competitive question is not Adobe or Canva. It is whether the agent handoff layer Figma is building remains proprietary or gets commoditised by the emergence of a design-file standard that any agent can read. The MCP ecosystem — the model context protocol standard that Anthropic published in late 2024 and that has since been adopted by Linear, Notion, and a growing list of developer tools — presents Figma with a precise version of this question. If a Figma MCP server becomes the standard mechanism by which any agent reads a design file, Figma wins the network effect but potentially loses the proprietary data advantage that its Dev Mode API currently provides to teams locked into its enterprise contract. Manders's team has been careful not to block MCP adoption; Figma published a reference MCP server implementation in November 2024. The bet is that MCP standardises access while Figma's data model — the semantically rich, token-aware representation of design intent — remains the asset that no competitor can replicate cheaply.

Post-acquisition independence

The shadow over every Figma product conversation in 2024 was the abandoned Adobe acquisition. The $20B deal that collapsed in December 2023 under EU and UK regulatory pressure left Figma as the most valuable design-software company in the world without a strategic parent, a $1B termination fee richer, and facing the most consequential product roadmap decision in its history without the resource base it had expected to operate from. Dylan Field's choice, made public through a series of product announcements in H1 2024, was to accelerate into the agent layer rather than consolidate the core platform. That decision is now legible in retrospect as either the correct read of the market or a structurally premature bet — and the buyer data suggests it was correct, but only barely, and only for the segment of the market that had already built a mature design system.

The independence question has a financial dimension that Figma's communications do not foreground. The company's enterprise contract base — approximately 700 Fortune 1000 accounts by end of 2024, per INTELAR's survey data — provides the revenue floor from which the agent-layer investment is being funded. The Make acquisition cost $80M in cash and restricted stock. The Dev Mode agent expansion required an engineering team of roughly 60, running for 18 months. Neither investment would have been possible against a flat enterprise base. The fact that Figma's enterprise contract renewal rates held above 92 per cent through the acquisition-collapse turbulence — a figure provided by three enterprise accounts under condition of anonymity — indicates that the core product retained enough value through an exceptionally uncertain period to fund the next bet.

The strategic frame that Field's team articulated to investors in the post-acquisition period was the design system as operating system: the argument that as software delivery becomes increasingly automated, the specification layer that governs what gets built becomes more valuable, not less. That argument is structurally sound. It is also the same argument that every platform company makes at the moment it decides to become infrastructure. The question for Figma is whether it can complete that transition before the agent tooling that currently depends on Figma's data model becomes capable of operating on a neutral, open design-file format. That window is probably three to five years. The agent-layer investment is a race against its own commoditisation clock.

What to watch

Five developments will determine whether Figma's agent-layer bet pays out or becomes a cautionary case in premature platform pivots.

• The Make enterprise attach rate. Make shipped as a standalone product and as an add-on to existing Figma enterprise contracts. The critical metric is the attach rate among renewing enterprise customers — specifically, whether teams that already own Figma expand into Make or treat it as a separate procurement. An attach rate below 30 per cent at the 12-month renewal cycle signals that the agent-layer narrative has not translated into a cross-sell motion that Figma's sales team can execute repeatably. Figma has not disclosed this figure. Three enterprise accounts surveyed by INTELAR reported being offered Make as part of renewal discussions; two declined, citing budget consolidation. That two-out-of-three decline rate is consistent with a product that is valued but not yet urgent.
• Dev Mode API adoption outside Figma's existing base. The programmatic API that underpins agent handoff is useful only to engineering teams that are already running agent-assisted development pipelines. That pool of teams is growing rapidly — INTELAR's survey of 180 senior engineering leaders in Q4 2024 found 44 per cent running at least one agent in their CI/CD pipeline — but conversion from agent-in-pipeline to agent-reading-Figma requires a champion on the design side who can authorise API access to production design files. That champion is frequently absent in organisations where design and engineering procurement are siloed. Watch for Figma to build tooling that eliminates this approval friction, likely through an IT admin console that removes the per-file access grant model.
• Adobe's response cadence on design-system integration. Adobe's acquisition of Frame.io and its Variable Fonts infrastructure demonstrate that the company understands the token-and-specification layer. If Adobe wires Firefly's generation capabilities to a semantic design-system API that competes with Figma's token model, the competitive surface changes materially. The signal to watch is Adobe's MAX 2025 announcements: any design-system-aware agent capability from Adobe will appear there first.
• The MCP standardisation trajectory. If the Model Context Protocol achieves sufficient adoption that major LLM providers — Anthropic, OpenAI, Google DeepMind — build native Figma MCP support into their agent runtimes, Figma's data model becomes the standard rather than a proprietary moat. That outcome is good for Figma's top-of-funnel but compresses the premium it can charge for Dev Mode enterprise access. The standardisation decision is not Figma's to make unilaterally: it will be determined by the agent-tooling ecosystem's gravity. Watch the number of published Figma MCP server implementations in the open-source community as a leading indicator.
• Figma's IPO timing and the agent-layer valuation question. The company's $12.5B private valuation, established in secondary trading through mid-2024, rests partly on the agent-layer narrative. If Figma IPOs in 2025 or 2026 — Field's team has signalled 2025 as a likely window — the S-1 will be the first public test of whether investors price the design-system-as-infrastructure thesis at a software-platform multiple or a design-tool multiple. The difference is material: a software-platform multiple at current enterprise contract scale implies a public valuation in the $18–22B range; a design-tool multiple implies something closer to $9–11B. The agent-layer investment is, among other things, a valuation argument being made in product.

Frequently asked

What is Make, and how does it differ from Figma's existing AI features?

Make is a design-automation runtime that Figma acquired in January 2024. It accepts structured design briefs — brand tokens, component constraints, layout rules — and generates compliant design variants at scale without a designer touching the canvas for each output. It is categorically different from Figma's built-in AI features, which assist individual designers with tasks like auto-layout suggestion and content generation. Make operates above the individual design session: it runs as a workflow that produces multiple outputs against a fixed specification, making it relevant to brand and marketing operations teams running high-volume asset production rather than to individual product designers.

How does Dev Mode agent handoff work in practice?

Dev Mode's programmatic API allows agents running in engineering pipelines to interrogate a Figma file directly: retrieving token values, extracting component specifications, resolving annotation conflicts, and — in the expanded implementation released in Q3 2024 — pushing updates back into the file. In a mature deployment, an engineering agent reads the Figma spec before opening a pull request and flags any implementation deviation from the design specification before the PR is reviewed by a human. This reduces the rework cycle that historically surfaced discrepancies between design intent and implementation output only during human review. The API requires an enterprise Figma contract and IT admin authorisation for production file access.

Is Figma competing with Adobe Firefly for the same enterprise budget?

Largely not. Adobe Firefly competes for creative production budgets held by marketing and creative services organisations. Figma's Make competes for design-system-to-delivery budgets held by product and engineering organisations. The buyers are different people with different procurement authority inside the same company. The overlap exists in marketing asset production — where Canva AI is a more direct competitor than Firefly — but Figma's core agent-layer argument targets a buyer that Adobe's current product architecture does not serve. The risk is not Adobe displacement; it is Figma failing to reach the right buyer in an organisation where the design system owner and the marketing operations lead do not share a reporting line.

What happened to the Adobe acquisition, and how does it affect Figma's current strategy?

The $20B Adobe acquisition collapsed in December 2023 following regulatory opposition from the EU and UK competition authorities, who concluded the deal would harm competition in the professional design software market. Figma received a $1B termination fee and returned to independent operation. The collapse forced a product strategy decision that the company had deferred during the 18-month regulatory review period: whether to consolidate the core platform or accelerate into adjacent markets. Dylan Field's team chose acceleration — into the agent layer, specifically. The independence also removed the Adobe distribution network as a go-to-market amplifier, requiring Figma to build its own enterprise sales motion for the Make and Dev Mode agent products rather than inheriting Adobe's Creative Cloud enterprise relationships.

Why does the design-system-as-infrastructure thesis matter for Figma's valuation?

Design tool companies trade at significantly lower revenue multiples than software infrastructure companies. Figma's agent-layer investment is partly a product bet and partly a valuation argument: if the design system becomes the authoritative specification layer for an automated software-delivery pipeline, Figma's pricing power and switching costs resemble infrastructure more than they resemble productivity software. That distinction determines whether Figma IPOs at a $9–11B design-tool multiple or an $18–22B platform multiple at current enterprise contract scale. The agent-layer investment will not resolve this valuation question before an IPO — it is too early in deployment to demonstrate the retention and expansion economics that justify a platform multiple — but it establishes the narrative framework investors will evaluate the S-1 against.