Not long ago, content used to be something you published. Now it's something your systems run on.

Today, when an AI agent answers a customer's question, it isn’t reading your website; it's reaching for whatever context it can find and acting on it. When automation updates your pricing across channels, AI is executing against your business logic, not browsing your CMS. When a support agent explains your return policy, it's not summarizing a help article; it's making a commitment on your behalf.

The world we knew has changed. Content has become context. And context has become infrastructure.

Andrej Karpathy said English has become the hottest new programming language. He's right, but the implications run deeper than most people think. The work isn't just happening in prompts or chats with agents. AI is drafting content on your site. Agents are executing workflows inside your organization. Assistants are acting on behalf of your customers, whether you’re aware of it or not. And every single agent is hungry for context in ways we never anticipated.

This is the shift, content is no longer something your audience reads, it’s become something your systems and workflows run on. It’s now infrastructure.

But where will the context come from?

You’ve heard this before. Just connect your Google Drive, your Notion workspace, and your internal wiki, and let the LLM agents figure it out. RAG will handle the rest.

It won't.

We keep seeing the same story. A team launches an AI assistant, then it starts hallucinating. They try better prompts, more retrieval, bigger context windows, but nothing sticks. Eventually, someone realizes the problem was never the model; it was that their product information lived in 47 different documents, none authoritative, half outdated, some actively contradicting each other.

The model wasn’t broken. The context was.

Here’s what keeps getting missed. Agents don't just need access to information. They need precise context they can act on. There's a difference between handing someone a pile of papers and giving them an API.

“I want a cheap winter coat.” Unstructured content means guessing. With embeddings and semantic search, you get "documents that seem related to affordable winter outerwear." That's useful for research. It's not acceptable when an agent is quoting a price or executing a transaction.

Structured content means knowing. You can query price < 200 AND available == true and get a deterministic answer. The agent isn't interpreting. It's operating.

The same is true if you’re in any industry that requires precise knowledge. If you're a travel company, pricing cannot be probabilistic. If you're a retailer, return policies cannot be "probably this." Structure gives agents facts to act on. Embeddings give them guesses to interpret.

So what does this actually look like?

Start with the foundation. Everything else depends on it.

At the center of this is your system of record, where content and business logic are modeled clearly and deliberately. What you sell, how things relate, what rules apply. Not scattered across wikis and shared drives, not buried in paragraphs that someone has to interpret. Something closer to a database than a document. Something both a human and a machine can reach into and trust what they find.

Around that foundation, humans and agents work together. Agents handle the routine: creating, updating, transforming content at scale. Humans handle the exceptions, the decisions that actually matter. This isn't about automating everything. It's about building infrastructure that supports both, and letting your organization grow into it at its own pace.

Governance doesn't disappear in this world; it gets more important. When humans, AI agents, and third-party systems are all touching your content and business information, you need to know exactly who changed what, when, and why. Whether it was a person or a pipeline. Permissions and auditability are what let you actually trust the system as it scales.

And it can't be an island. Your content infrastructure has to connect outward, feeding context to external agents like Claude or ChatGPT, pulling in capabilities from tools like Shopify or Linear, distributing not just to websites and apps but through APIs and MCP servers that are quickly becoming just as important as any traditional publishing channel.

You don't have to rebuild everything at once. Most teams start with one use case: a website, an agent, automating content workflows, or a customer service backend. But here's what we keep seeing: each new use case makes the whole system more valuable. It compounds. The thing that started as a website backend quietly becomes the system of record for the entire organization.

That's not a side effect. That's the point.

The AI Content Operating System

This is the architecture that's emerging. Not a CMS with AI features bolted on. Not a vector database hoping semantic similarity is close enough. Something purpose-built for the world where content requires structure, where humans and AI collaborate, where content powers intelligence.

We call it The AI Content Operating System. A content foundation at the core, where content is more than text and images, it’s business logic, and it’s structured by design. Custom interfaces for humans and agents to work together, with governance built in. Connections to the rest of your stack and to the broader ecosystem of AI assistants and tools. The ability to power any application, from websites to agents to agentic workflows, from a single governed source, to give you full control.

Could you build this yourself? Sure, wire up a database, add some APIs, call it a content system. But then you've built infrastructure with no content semantics, no governance layer, no way for agents to operate reliably, no interfaces for humans to interact with AI seamlessly. And you'll be maintaining it forever, you know the story.

The companies that have built their foundation in structured content to automate content operations have an advantage over their competition. Every new channel, every new agent, every new automation draws from the same foundation. The ones that haven’t will spend years reconciling silos, fixing hallucinations, apologizing when agents get it wrong. The differentiator is your ability to innovate, to grow your business.

This is what we've been building at Sanity — The AI Content Operating System, with humans in control and AI accelerating both content as a strategic asset and business velocity.

Content has become context. Structure powers intelligence.

The question is whether you're building that structure or cleaning up after agents that guessed.

Learn more about Sanity Agent Context.