The Assistant Recommends You and Replaces You. Canva's Valuers Just Priced It.

Canva has 265 million monthly active users, nine years of profitability, and $921.9M in Q2 revenue, up 25.2% year over year. In August 2026, its investors marked it down anyway. Blackbird and AirTree cut their valuation from $42B to $34.9B, a 17% reduction. A second investor marked it down 20% to $31B, from $38.9B a year earlier. The Hiive secondary market implies roughly $30B, a 29% discount. Three independent marks, all pointing the same direction, on a company whose operating metrics are excellent.

The reason is in one sentence from SaaS investor Jason Lemkin, quoted in the State of Brand analysis: “Agents just routed us around Canva.”

We have written about how volatile assistant recommendations are, and about where agents go when a vendor’s pricing page fails them. Those pieces covered visibility: whether the assistant mentions you at all. This is different. This is the first time the market has attached a valuation number to the net effect of assistants on a specific company, and it happened to a company the assistant channel visibly feeds.

The Same Assistant, Both Directions

Look at Canva’s own AI numbers, from the same State of Brand piece. Users have had 26 million ChatGPT conversations with Canva’s app. The product logs 800 million AI tool uses per month. By any distribution metric, the assistant channel is working. ChatGPT sends Canva users; Canva’s own AI features are heavily adopted.

And yet the marks went down. Because the same assistant that opens a Canva session for one user generates the social graphic directly for another. The assistant is simultaneously the company’s largest new referral channel and its most credible substitute. Every conversation is a fork: route the user into the product, or absorb the job the product exists to do.

There is no dashboard that nets these two flows against each other, which is precisely why the valuation marks matter. Growth-stage investors do not need the dashboard. They price the expected net, and three of them independently priced it negative to the tune of 17% to 29%, against 25% revenue growth. The market is saying that the substitution flow is growing faster than the referral flow, or will be soon.

Relief Brands and Identity Brands

The State of Brand piece, an editorial analysis without a byline, offers a frame for which companies this hits hardest. It splits loved brands into two kinds. A “relief” brand is loved out of gratitude: it solved a painful problem, and users are thankful. An “identity” brand is loved because using it says something about who the user is.

Canva is the canonical relief brand. Non-designers loved it because it removed the pain of producing a decent graphic. But gratitude for a solved problem has a structural weakness: when a new technology removes the problem itself, the gratitude transfers to the new technology. If an assistant produces the graphic inside the conversation, the relief now lives in the assistant. The 265 million users do not have to churn angrily. They just have one less reason to open the tab.

Identity brands are exposed differently. The assistant can replicate the function, but it cannot replicate what the brand membership signals. Per the State of Brand analysis, that is the layer of brand value that survives the routing.

For anyone operating a software business, the uncomfortable exercise is to ask which kind of love your revenue sits on. If your NPS is high because you removed a pain, an assistant that removes the same pain inherits your customers by default.

Fifty Vendors, Four Answers

If the substitution threat is real and now priced, you would expect vendors to be answering it in their positioning. Kyle Poyar’s audit at Growth Unhinged, published in August 2026, measured exactly that. He examined 50 SaaS and AI-native companies for how they handle the fact that a prospect’s realistic alternative is no longer a competitor but building the workflow with Claude directly.

The results, from a July 2026 review:

34 of 50 shipped MCP servers or connectors, plumbing that lets models route through their product instead of around it.
12 name Claude, the actual alternative, anywhere on their own domain.
4 publish a build-versus-buy cost comparison a prospect could use.
18 publish nothing on the subject at all.

Read those numbers as a sequence. Most vendors have made the infrastructure move: the MCP server is the technical answer to routing, an attempt to become the surface the agent lands on rather than the surface it skips. Far fewer will say the competitor’s name. Almost none will engage with the buyer’s real question, which is what the do-it-with-the-model path actually costs in engineering time, maintenance, and reliability.

That last silence is the strange one, because the build-vs-buy comparison is the one artifact where an incumbent vendor holds every advantage. The vendor knows its own uptime engineering, its edge cases, its support load. A prospect prompting an assistant to estimate the same thing will likely get a generic answer. Forty-six of fifty vendors are leaving that argument unmade, on their own domain, while the buyer has the conversation with the assistant anyway.

The Paradox Is a Budget Line Now

The two sources stack into one conclusion. State of Brand shows the cost of not answering: three valuation marks down between 17% and 29% on a company with strong fundamentals, justified by a single routing sentence. Growth Unhinged shows that the vendor population mostly has not started answering: infrastructure shipped, argument absent.

Our 500-run study coverage showed agents falling back to third-party sources when vendor sites fail them. Combine that with Poyar’s 18-of-50 silent vendors and the picture sharpens: when the assistant or the buyer asks “should we just build this with Claude,” the answer is being composed from everything except the vendor’s own materials. The vendor’s absence from that conversation does not pause the conversation.

The Canva marks convert this from a marketing concern into a finance one. Once one high-profile company has been repriced on agent routing, every growth-stage investor has a precedent to apply to the next portfolio review. The question “what fraction of your value proposition survives an assistant doing the job” now comes with a precedent attached, so expect it in the next diligence round. Companies will be answering it either in their own positioning or in someone else’s spreadsheet.

Do This Now

Run the two-sided audit on your own product this week.

Side one, the substitution ledger: take your ten most-used features and ask, honestly, which of them a frontier assistant can perform end to end inside a conversation today. For each one it can, classify why customers would still come to you: workflow depth, data they have already stored with you, compliance surface, or identity. If the honest answer is “habit,” write that down too. That column is your repricing exposure.

Side two, the positioning check, straight from Poyar’s four measurements: do you have an MCP server or connector, do you name the assistant as an alternative anywhere on your domain, and do you publish a build-vs-buy comparison with real numbers for engineering cost, maintenance, and reliability? If you are in the 18-of-50 that publish nothing, the assistant conversation about your category is happening without you, and the Canva marks show what the market does with that silence.

This analysis synthesizes Canva: Loved Brand, Valuation Markdown (State of Brand, August 2026) and Meet Your Biggest Competitor (Claude) (Kyle Poyar, Growth Unhinged, August 2026).

Victorino Group helps software companies audit their exposure to assistant substitution and build the positioning and integration surface that keeps agents routing through the product. Let’s talk.