AWS is helping vibe-coding startup Superblocks, and the implications are big
TechCrunch AI Julie Bort
AWS just struck a multi-year deal to embed Superblocks' vibe-coding tool inside enterprise private clouds. That means business users get AI app-building without data ever leaving the company's AWS account.
Superblocks makes vibe-coding tools that let non-engineers describe an app in plain English and watch it get built. Now, thanks to a new multi-year marketing agreement, that tool is getting wired directly into the private clouds of AWS customers. No data leaves the building. No app gets spun up on some external database nobody in IT has heard of. Instead, everything runs on Amazon Aurora inside the customer's own AWS account, hooked into Bedrock for the AI plumbing, and covered by whatever encryption, auditing, and network controls that enterprise already has in place.
CEO Brad Menezes frames it simply: bring vibe coding to the data, not the other way around. That distinction matters more than it sounds. Plenty of vibe-coding startups have thrived by being fast and loose, letting business users spin up apps that live outside IT's purview entirely, often on tools like Supabase. Superblocks is betting the opposite instinct wins in the enterprise: compliance officers and CISOs want these tools, they just want them leashed.
AWS gets something out of this too, beyond a cut from Marketplace sales. It doesn't have its own vibe-coding product for business users yet. Kiro is aimed at developers. Quick is more of a Copilot-style assistant. So partnering with a 50-person startup that's raised $60 million lets Amazon claim a foothold in a category it hasn't built internally, while keeping the whole workflow — model calls, data storage, orchestration — inside its own cloud.
That's really the bigger story here. Satya Nadella has spent weeks telling Microsoft's enterprise customers not to hand their AI scaffolding to model labs, warning that firms like OpenAI or Anthropic could study a company's data and eventually compete with it. Enterprises seem to be arriving at the same conclusion on their own. Menezes says the model loyalty that existed even two months ago has evaporated — companies that once insisted on a single provider, often Anthropic, are now demanding multi-model setups that include Chinese open-weight models alongside the usual frontier names. Vercel's own numbers back this up: open models made up 29% of traffic through its AI gateway last month.
Once enterprises refuse to lock themselves to one model, they can't lock their orchestration layer to one provider either. That's the opening AWS and Microsoft are racing to fill, and Superblocks just became AWS's chosen vehicle for doing it in the vibe-coding lane specifically. Menezes doesn't mince words about where this is headed, predicting any exec who bets an enterprise on a single model provider will eventually get fired for it.
My take
This is the cloud giants doing what cloud giants always do: watch a category get proven risky and chaotic by scrappy startups, then swoop in to sell the tamed, compliant version to enterprises terrified of shadow IT. Nothing wrong with that, but let's not pretend AWS suddenly loves vibe coding — it loves owning the data pipe and taking a Marketplace cut. The real signal is that multi-model, provider-agnostic infrastructure just became the default enterprise posture, and the AI labs' dream of being the whole stack is dying fast.
Read more about this at: TechCrunch AI