“Every few months, a new model made part of our roadmap unnecessary”: Why Mendral’s founders gave up their startup for Anthropic
The New Stack Amanda Caswell ● Covered by 3 sources
Anthropic just acqui-hired Mendral, a YC startup building AI agents for CI/CD grunt work like flaky tests and security checks. The founders admit Claude's own progress kept eating their startup roadmap, so they joined Anthropic instead of competing with it.
Based on reporting by The New Stack, Amanda Caswell — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Sam Alba and Andrea Luzzardi know their way around the guts of software infrastructure. Alba ran engineering at Docker, and Luzzardi wrote some of Docker's earliest code back when it was still called dotCloud. So when the pair started Mendral, fresh out of Y Combinator's Winter 2026 batch, they weren't dabbling. They built three always-on agents: one that hunts for leaked secrets and locks down dependency versions, one that patches flaky tests, and one that shaves time off builds through smarter caching and parallelism. All of it ran inside Blaxel's sandboxes, isolated compute environments built for agents rather than humans, with resume times under 25 milliseconds.
Here's the twist. Mendral leaned on Claude from its very first commit, and the founders say that dependency eventually swallowed the company whole. Every time Anthropic shipped a new model, a chunk of Mendral's planned roadmap became pointless, made obsolete by capabilities the model now handled natively, while an even bigger chunk of new work suddenly became feasible. That's not a comfortable position for a startup trying to carve out a defensible product. It's a good position, though, if you're the lab building the models.
So instead of continuing to build a business shadowing Anthropic's release schedule, Alba and Luzzardi are joining Anthropic outright. Mendral's hosted product is winding down, existing customers are getting help transitioning elsewhere, and neither side has disclosed what changed hands financially. The founders framed it plainly in their announcement: there's no better place to work on what software engineering is turning into than inside the company setting the pace of that change.
This isn't Anthropic's first grab of infrastructure talent this year. Back in the spring, it picked up Stainless, a startup known for generating SDKs and building out Model Context Protocol tooling. Line those two moves up and a pattern appears. Anthropic isn't just chasing bigger, flashier models anymore. It's assembling the boring-but-essential plumbing, secrets scanning, dependency management, SDK generation, that enterprise engineering teams actually live inside every day. Claude the chatbot is one thing. Claude wired into a company's entire CI/CD pipeline is a different, much stickier business.
My take — AI-written commentary, not fact-checked reporting
This is the tell nobody wants to say out loud: if your startup's roadmap keeps getting vaporized by someone else's model update, you don't have a moat, you have a to-do list for a bigger lab. Mendral made the rational call, but it's a warning shot for every thin agent wrapper out there building on top of Claude or GPT instead of underneath them. Anthropic, meanwhile, keeps quietly buying the unglamorous plumbing that makes enterprises sticky, which tells you more about its long-term ambitions than any benchmark chart does.
Read more about this at: The New Stack