Mailchimp is shrinking inside Intuit, and the named cause is that agents can't reach it

Jason Lemkin traced the numbers: Mailchimp peaked around $1.06B ARR in Q4 FY22, the only quarter Intuit ever disclosed the figure, and by Q3 FY26 it's in outright year-over-year revenue decline. Intuit paid $12B for it in 2021 when it was doing $800M and growing ~20%. Intuit has now reported growth rates excluding Mailchimp for four straight quarters, with the gap widening from half a point in September 2025 to three points plus a decline by May 2026. There was a 17% workforce cut, roughly 3,100 roles, and $300M in restructuring charges.

Klaviyo over the same window: $358M in Q1 2026, up 28%, 110% NRR, 196,000+ customers.

The cause Lemkin names is the one that should make every SaaS founder reading this open a new tab. Mailchimp has no official MCP server for its marketing features. It's absent from Replit, Lovable, and Vercel, which is where new applications now get assembled. Resend is wired into all of them.

I want to be careful here, because "no MCP server" is not a complete explanation for a $12B acquisition going sideways. Post-acquisition integration friction, pricing changes, brand dilution, and a departing founding team all matter and all predate agent operability. Lemkin's diagnosis is the marginal cause, not the whole cause. But it's the marginal cause that's still compounding, and the other ones already happened.

Here's the mechanism as I understand it. When a developer spins up a new product in Lovable or Replit and needs transactional email, they don't comparison-shop. They describe what they want and the agent picks something it can actually wire up. Being reachable by an agent isn't a marketing channel anymore. It's the distribution channel for the entire cohort of software being built this way, and that cohort is growing.

The uncomfortable pairing: this story runs in exactly the opposite direction from the MCP governance story above. One is enterprises building allowlists to keep agents out. The other is a product dying because agents couldn't get in. Both are true simultaneously, and the resolution is that governed access is what wins. Not open access, not closed. An MCP server with real OAuth and real scoping is a product feature and a survival requirement at the same time.

Related data point on the same axis: Can I Vibecode It now rates 996 SaaS applications with a YES / KINDA / NOT REALLY verdict and a paste-ready prompt, up from 976 on August 6. Its "Death List" tracks roughly $18,827/mo in cancelled subscriptions across ~700 products, topped by Granola (181 votes), Wispr Flow (96), and SuperX (75). 209,383 visits since July 29. The composition is the signal: the most-replaced tier is thin-wrapper single-workflow tools, which is precisely where the last three years of seed funding went.

If you sell software: two questions. Can an agent operate your product end to end? And is your product more than one workflow? A no on either is a countdown.