Canva cut 1,271 jobs after users wanted more AI than it could price

The design-tool market just had a rough two weeks. Canva cut 22% of its staff. Squarespace cut 19%. And Figma kept shipping AI features that reach deeper into the work your team does every day. These things are connected. Let me catch you up.

The deep cut

• AI demand grows faster than AI economics. Canva's users wanted more AI than Canva could serve at a price that worked.
• Cheaper per task is not cheaper overall. Canva cut cost-per-task 90% and still slashed its growth forecast and 1,271 jobs.
• Specialized models beat frontier models on narrow jobs. Figma's own editing model matches or beats its old agent 62% of the time.

The bill came due for free AI

Canva's problem was not weak demand. It was the opposite. Perkins told Fortune that user appetite for AI features "significantly exceeded" the company's expectations. The trouble was math. Every AI request leaned on expensive frontier models, and a big share of that traffic came from free users. In August, Canva cut its revenue growth forecast from 30% to 20% to deal with the cost.

Then came the layoffs of 1,271 people, 22% of staff, in what Perkins called Canva's "hardest year yet." Investors had already cut its valuation to about $35 billion. The lesson for your roadmap: if you ship AI features your users love but cannot price, popularity becomes a liability, not a win.

Cutting cost-per-task is not the same as cutting cost

Canva did the engineering work. It built its own models to stop renting so much compute. Perkins said cost-per-task dropped 90% since Canva AI 2.0, and the newer Canva Brush cut it another 46%. Those are real numbers.

And it still was not enough. The forecast still dropped. The jobs still went. When usage climbs faster than your unit costs fall, the total bill keeps growing. Bring that framing to your next review. Ask what happens to your AI spend if adoption doubles, not what one task costs today.

The same squeeze hit website builders

Squarespace cut about 300 people, 19% of its workforce, a day before Canva's announcement. The pressure here came from a new crop of "vibe-coding" tools that spin up a working site from a short prompt. Lovable raised a $400 million round at a $13.3 billion valuation. Wix bought Base44. Google shipped Stitch.

Squarespace is pushing back with its own AI, including an upcoming platform called Forma. But a company that went public at $10 billion and then went private in a $7 billion deal is now cutting headcount to "move faster." The category you compete in can shift under you in a year. Watch who is raising money near you, not just who is shipping.

Figma's bet: train your own narrow model

While others cut, Figma is trying a different fix for the cost problem. Instead of routing everything through pricey frontier models, it trained a Figma-fluent editing model for the small edits designers repeat all day. Change a color. Fix spacing. Rework a layout after feedback.

The results back the approach. The agent beats or matches the old one 62% of the time, user ratings are 33% higher, and people undo its work 60% less often. Figma also more than doubled the share of work handled by its own model and dropped the average cost per message. The agent still calls third-party models for image generation and fresh explorations, and uses the in-house model for precise edits.

For your team, this is the playbook Canva is reaching for the hard way. Figma's code layers feature now lets designers clone a GitHub repo onto the canvas, generate three variations of a feature from one prompt, and hand a developer a spec. The tool is moving closer to production work, which changes what a designer's day looks like.

Three questions for your team

• If AI adoption on our product doubled next quarter, does our cost per user go up or down? Get a real number before your next planning cycle.
• Which repeated tasks in our workflow could run on a cheaper specialized model instead of a frontier one, the way Figma split its agent?
• As tools like code layers push designers toward code and specs, which roles on our team grow and which ones shrink? Decide before the market decides for you.