Sanity

1,000,000 users. One million people building on Sanity. Newsroom editors, developers, merchandisers, marketers, localization teams, product advisors. What they're shipping has changed. Alongside the websites and apps Sanity has always powered, teams are now running their own agents against the content layer. A merchandiser ships a product price change, and the agent updates every reference across eleven country sites in a single pass. A newsroom team points an agent at a decade of archives and asks where the coverage gaps are. A marketing team migrates thousands of unstructured web pages into a schema the editorial team can actually work in, overnight. A product team audits every help article for outdated screenshots and triages the ones that need redoing. None of that is content for a person to read on a page. It's content an agent can retrieve by meaning and write back as a versioned, permissioned document the editorial system can trust. That's the job Sanity was built for. Thanks to everyone shipping on it. You're why this number exists.

I've been complaining about the lack of decent CMS options for ages. Having started out in Marketing on the web, I've worked with most of the major ones. What I've found is that the ones with all the buzz often suck for anyone actually doing any web publishing at scale. But rolling your own CMS is also, as Lovable rightfully concluded, "overkill". When I asked what I should use for a CMS for the Sfumo blog I'm building, Lovable recommended Sanity. I had heard of the brand but hadn't explored it, and...I'm sold. Feels like I'm living in the future. Nerding out levels are unmatched. I'm not the core target audience (developers), but Lovable does everything for me via MCP. Every time it changes schema, it gives me instructions for what to change locally (which I just pass to Claude Code to do), and how to redeploy via the terminal (3 words). And I get a custom-built WYSIWYG web editor for minor copy tweaks I want to make manually. So far, here are the tools I'm using: Lovable - design/webdev Gamma - images (their new offering is so far my favorite image generation tool. Nails brand look and feel.) Sanity - CMS that's agentic but also friendly to non-developers Claude Code - editing .ts files and manipulating images (removing BG, changing perspectives) I'm on the third iteration of the design and am having way too much fun. Finally feels like the tools are getting to the point where they can keep up with my imagination. Next up: with the design, CMS, and scaffolding in place, now I have to actually get to writing! :partyparrot:

What if your AI agent could query your content with typed filters, traverse relationships, and rank by semantic relevance, all in one expression? Sanity now has built-in embeddings on every plan. You can enable it on your existing dataset with a single CLI command. GROQ lets you combine structured filters with semantic search in one expression. Agent Context exposes all of it to any MCP-compatible agent as a hosted endpoint. Your content is already ready. No rework, no re-indexing. Install the Agent Context skill to get started in minutes: `npx skills add sanity-io/agent-context --skill create-agent-with-sanity-context` And tell your coding agent: "Use the create-agent-with-sanity-context skill to help me build an agent." You should have a working prototype within 5 minutes! Works with Claude Code, Cursor, and any editor that supports agent skills.

"It's becoming much more available without a really heavy engineering lift" Andy Fitzgerald, PhD, Content Operations Consultant (Elemeno Health / World Health Organization), has been experimenting with Skills to tackle what he calls "non-human-shaped problems," building taxonomy systems grounded in open standards, and testing LLMs against professional taxonomists for content tagging. Last month at our Sanity Pioneers SF event, he was joined by Meaza Abate, Engineering Manager at Airtable, sharing how they use AI in their content operations day to day. Watch the full conversation on our YouTube.

Content teams aren’t blocked by publishing. They’re buried upstream - rewrites, translations, reuse that never works. And now content isn’t just for people anymore. It feeds models, powers agents, run workflows. Content needs to be data: structured, governed, reusable. Sanity is the intelligent backend for companies building AI content operations at scale. Model your business. Automate everything. Power anything. Structure powers intelligence.

"We're using AI agents to evaluate the quality of the alt text" At our Pioneers event in SF, Evelina Wahlström sat down with two practitioners who've spent years building content operations workflows with Sanity. Meaza Abate, Engineering Manager at Airtable (ex-Figma), has built content ops workflows ranging from localization pipelines to migrating content using the Sanity MCP in weeks rather than months. She was joined by Andy Fitzgerald, PhD, Content Operations Consultant working with Elemeno Health and World Health Organization, who gets into taxonomy design, domain modeling, and experimenting with Skills for problems he says "aren't human-shaped." Watch the full fireside chat on YouTube.

I've been on a meal plan since last winter. Lost 25 lbs, which is great. But I'm the one who cooks, and managing a strict plan for a whole family is hard. So I built an app. PDFs from my nutritional coach go in, structured recipes come out. Scaling for my wife and kids, calculated weights, search powered by Algolia. One thing was missing: pictures. My kids would scroll through and just see a list of text. Not exactly inspiring when you're trying to pick dinner. Then I saw that Sanity shipped image generation in the Agent API. It reads your schema and uses your actual content to drive the prompt. "Coconut and oat pancakes with blueberries" gets a photo of that dish, not a generic breakfast stock image. I ran it on 50 recipes over a weekend. The kids actually want to pick dinner now. Wrote up the full process on the Sanity blog, including the complete TypeScript script and a couple of gotchas that tripped me up (arrays and the draft/publish flow). Here's the full write-up: https://lnkd.in/gUduGdbC

For the longest time, CMSes have stored rich text as HTML or Markdown. It works for rendering a webpage. It falls apart when you need that same content in a mobile app, a PDF, a search index, or an AI pipeline. You end up parsing markup instead of working with data. Not fun! Portable Text is an open specification for block content as structured JSON. It's also an editor for authoring that content. Two projects, one ecosystem. The spec: → Block content stored as an array of typed JSON objects. Text, images, code blocks, or any custom type you define → Annotations and inline objects carry structured data, not markup → Queryable. Filter by block type, extract all links, transform for any output → Render in any language. Serializer packages for React, Vue, Svelte, Astro, HTML, Markdown, plus community libraries in C#, Python, PHP, Ruby, Go, Flutter, and Hugo The editor: → Schema-driven. Define your content model, the editor enforces it → Headless. No built-in UI. Full control over toolbar, rendering, and interaction → Behavior API for custom keyboard shortcuts, paste handling, and input rules → Nine official plugins. Testing infrastructure with Gherkin specs and Vitest Portable Text has been in production since 2017. It powers Sanity Studio and Canvas and has been adopted by projects like Hugo (built into the binary). 50+ packages across eight languages. We just shipped a major docs overhaul. Per-framework rendering guides, custom blocks walkthrough, conversion to/from HTML/Markdown, editor behavior recipes, and an ecosystem catalog covering every package and plugin we could find. We're currently working on improving support for editing tables and more complex block types. You can build them today with custom blocks, but we want to make it cleaner. https://lnkd.in/g5JBzuC7