On February 17, a widely circulated analysis argued that LLMs are "dismantling the moats that made vertical SaaS defensible" — that domain-specific software companies are losing their advantage as AI can replicate their specialized workflows. The same day, Figma and Anthropic announced Code to Canvas, a feature that lets users import code into Figma using Claude Code. The next day, Figma reported Q4 revenue of $303.8 million, up 40% year-over-year. Canva's COO disclosed the company hit $4 billion in annual recurring revenue at the end of 2025. One essay declared the moats are collapsing. Two companies reported the moats are growing.
Both are true. The question is why.
The SaaS Moat Thesis
The argument against vertical SaaS goes like this: software companies built defensible positions by encoding domain expertise into workflows. A legal tech company knows how lawyers review contracts. A healthcare platform knows how hospitals process claims. That domain knowledge — captured in UI patterns, integrations, and data schemas — was the moat.
LLMs dissolve that moat because they can learn domain expertise from context. An AI agent with access to a company's data can perform the same workflows that required specialized software. The domain expertise moves from the software to the model. The vertical SaaS company becomes a thin UI layer over capabilities that any LLM can replicate.
This is a reasonable thesis. And it's already happening in some categories — customer support, data entry, basic legal review. But it misses something about what certain tools actually do.
What Figma Actually Sells
Figma doesn't sell design automation. It sells a collaborative canvas where design decisions are made, shared, and iterated. The product is the workspace, not the workflow. When Code to Canvas lets users import code into Figma via Claude Code, it doesn't replace Figma. It makes Figma the destination for AI-generated output.
This is the crucial distinction. Some SaaS products automate tasks. Others host collaboration. Task automation is vulnerable to AI substitution — if the task can be described in a prompt, the software is redundant. Collaboration hosting is not vulnerable, because the value is in the shared context, not the execution.
The Anthropic Partnership
The Figma-Anthropic partnership is revealing. Code to Canvas means a developer can write code, and Claude translates it into a Figma design that non-technical teammates can review and modify. The AI isn't replacing the design tool. It's feeding the design tool.
This inverts the standard AI-disruption narrative. The fear is that AI makes software tools obsolete. What's actually happening, at least in design, is that AI makes the tools more valuable by expanding who can contribute to them. A developer who couldn't use Figma before now generates designs programmatically. The design team still reviews, iterates, and approves in Figma. The collaboration surface grows.
The Productivity Data
There's a macro signal hiding in the same week's news. The Financial Times reported that US productivity rose approximately 2.7% in 2025, nearly doubling the 1.4% average of the previous decade. The productivity gains are real. The question is where they're showing up.
If AI were primarily destroying software moats, you'd expect to see SaaS revenue growth decelerating as customers replaced specialized tools with general-purpose AI. Some categories are seeing this. But the design tools — the visual collaboration platforms — are accelerating. Figma's 40% growth is faster than most of its cohort. Canva's $4 billion in ARR came during the same period that AI tools were supposed to make design templates obsolete.
The productivity gains from AI are flowing through these tools, not around them. Designers are more productive because of AI, and they're being productive inside Figma and Canva.
Which Moats Collapse
The LLM-moat thesis is correct for software that automates deterministic tasks. If your product's value proposition is "we know how to do X and we encoded it in software," an LLM that can also do X makes your software less valuable.
The thesis is wrong for software that hosts creative collaboration. These tools succeed not because they automate work but because they provide a shared surface where humans make decisions together. AI makes the inputs to that surface richer and more accessible. It doesn't replace the surface itself.
The moats that are collapsing are workflow moats — the ones built on "we know the steps." The moats that are holding are context moats — the ones built on "everyone works here." Figma's moat isn't that it knows how to design. It's that every designer's team is already in Figma. Code to Canvas makes that moat deeper, not shallower.
The tools that host collaboration get more valuable as AI makes more people capable of contributing. The tools that automate tasks get less valuable as AI makes the tasks trivial.
The SaaS reckoning is real. But it's a sorting, not a collapse. The question for every software company isn't "will AI replace us?" It's "are we a workflow or a workspace?" This week's numbers suggest the workspaces are winning.