Cursor Ditched Sanity and Built Their Own CMS in 3 Days. Is This the Future—Or a Trap?
Last week, Lee Robinson—former VP of Developer Experience at Vercel and now at Cursor—dropped a blog post that sent shockwaves through the developer community. He migrated cursor.com off their headless CMS to raw markdown files and code. It took 3 days. It cost $260 in AI tokens. He used 344 agent requests.
Sanity, the CMS he migrated away from, fired back with a 3,000-word response titled "You should never build a CMS."
This isn't just a story about one company's website migration. It's a glimpse into a fundamental question about the future of software: Will AI agents make SaaS infrastructure obsolete? Or is "vibe coding" your way to custom solutions a trap?
What Actually Happened
Here's Lee's account from his blog post:
"I was eating lunch with Roman and Eric at Cursor and we started talking about the cursor.com website. We recently shipped a redesign, and as part of that change, all content and web pages were now built through a headless CMS. We went around the table airing our grievances. The CMS wasn't working."
— Lee Robinson
The core frustration: AI agents couldn't easily access content behind authenticated APIs. When you can @cursor and ask it to modify code directly, having to click through CMS menus feels like a regression.
So Lee did what any engineer at an AI company would do: he asked the agents to fix it.
The Migration by Numbers
- Time: 3 days (estimated weeks)
- Cost: $260 in tokens
- Agent requests: 344
- Code removed: 332,000 lines
- Code added: 43,000 lines
The new setup: markdown files, GitHub for version control, Vercel for hosting, and a "vibe-coded" media management interface built with "3-4 prompts."
Sanity's Response: "You Just Built a CMS"
Sanity's Head of Developer Community, Knut Melvær, published a thoughtful rebuttal that deserves attention. His core argument:
"Read his post carefully and look at what he ended up with: An asset management GUI, user management via GitHub permissions, version control via git, localization tooling, a content model (markdown frontmatter with specific fields). These are CMS features. Distributed across npm scripts, GitHub's permission system, and Vercel's infrastructure."
— Knut Melvær, Sanity
In other words: You can delete the CMS, but you can't delete the problems a CMS solves.
The Case for Vibe Coding Your Infrastructure
Lee's argument has merit, especially for teams like Cursor:
1. AI Agents Work Better with Code
When your content lives in markdown files, agents can grep it, modify it, and reason about it directly. No API authentication. No SDK. No context-switching between systems.
2. Reduced Complexity
332K lines of code became 43K. That's not just fewer lines—it's fewer abstractions, fewer dependencies, fewer things that can break.
3. Speed
Prefect, another company, saw Lee's post and shipped their own response in 34 minutes. The bottleneck isn't coding anymore—it's having systems where agents can operate.
The Case Against: What Breaks at Scale
Sanity's counterarguments are equally compelling:
1. The Content === Page Trap
One markdown file per page works until your pricing appears in three places. Then you're either duplicating content (maintenance nightmare) or building a templating system that pulls from canonical sources. At which point... you've invented content references. You're building a CMS.
2. Git is Not a Content Collaboration Tool
Git was built for code, which has different properties than content:
- Code merge conflicts are mechanical. Content merge conflicts are semantic.
- Branching maps to features in code. "I'm working on the Q3 campaign" isn't a branch—it's 30 pieces of content across 12 pages with 4 people contributing.
- Real-time collaboration matters for content. Git is async by design.
The Deeper Question: Will Vibe Coding Kill SaaS?
"Vibe coding"—a term coined by Andrej Karpathy—refers to using AI to generate code from natural language prompts rather than writing it manually. Lee's migration is a perfect example.
The implications are profound. If you can build a custom CMS in 3 days for $260, why pay for Sanity? Or Contentful? Or any SaaS tool?
The Bull Case for Vibe-Coded Infrastructure
- Perfect fit: Custom solutions match your exact needs
- No vendor lock-in: You own everything
- Agent-native: Built for AI from day one
- Cost: One-time $260 vs. recurring SaaS fees
The Bear Case
- Hidden complexity: The features will accrete because the problems are real
- Maintenance burden: Who fixes it when it breaks?
- Security: Vibe-coded solutions may have vulnerabilities
- Scale: What works for 50 developers doesn't work for 500
Who Wins?
Both sides have a point, and the answer depends on context:
Vibe Coding Wins When:
- Your team is small and technical
- Content has a single destination
- You ship infrequently
- You value agent-native workflows above all
- You're willing to maintain custom infrastructure
Structured CMS Wins When:
- Content needs to flow to multiple channels
- You need real-time collaboration
- Governance and compliance matter
- Your content team isn't all engineers
- You need content to be queryable, not just grep-able
The Real Lesson
Lee's post went viral because it resonated. Developers have real frustrations with tools that were built for a different era—an era before AI agents could write code for you.
But Sanity's response is equally important: you can delete the abstraction, but you can't delete the problem. Content management is complex because content is complex. Whether that complexity lives in a SaaS product or a folder of markdown files, it exists.
The companies that win will be the ones who figure out how to give AI agents first-class access to structured content—not the ones who flatten everything to strings so grep can find it.
Or maybe Lee's right, and in 6 months we'll all be vibe coding our infrastructure from scratch.
Either way, the experiment is running. And the AI code editors are watching.
Lee Robinson's full post is at leerob.com/agents. Sanity's response is at sanity.io/blog/you-should-never-build-a-cms.