Sanity

Every content team has a backlog of small fixes nobody gets to. The page that's out of date, the doc with a syntax error in a code block, the feedback that piles up. Durable agents are good at exactly that content operations work. We paired Sanity with Vercel's brand new eve framework and made an example of a docs agent that picks up the job, reads feedback for content, and writes a structured fix back as a draft for a human to ship. Because it runs on eve it's durable, so it reacts immediately to new feedback, retries (instead of dying on an error), resumes, and also runs on a schedule to pick up whatever wasn’t picked up. ↳ Starter: https://lnkd.in/ddM9pe-J ↳ Guide: https://lnkd.in/dsB9a-jC

If you've shipped content for real clients and teams, you already know "where should content live" stops being a plumbing question the moment it's more than a blog or a simple marketing site. It's a governance and operations problem. Keeping one fact consistent across a dozen places, many hands (with different permissions) on it at once, multiple releases that ship or roll back together, who's allowed to touch what. That's content ops, decades deep in hard problems. The "just keep it in markdown/embeddings and let agents handle it" instinct comes from circles that, I suspect, never had to live that and only focus on simple website publishing. I have tried it anyway on a couple of our own simple sites, and it started to break exactly where experience said it would, when the same content had to repeat across pages. Sure, I had the agent check for content drift every run (so much time spent scouring files), but why pay tokens and patience to imitate what a database solved ages ago? Flip it around and agents plus a real content database open up content ops like we haven't seen before. I'm unpacking what this means, and a lot more, in my new essay on the Sanity blog. https://lnkd.in/g3ibYxX4 What pains and gains have you noticed when you try to put agents on content ops tasks?

A system that reconciles itself while you're sleeping? That's Sanity Home. Our intranet. Saskia Bobinska built a set of scheduled functions that handle the boring manual work. Some run overnight, reconciling anything that's drifted, so the team wakes up to everything in order. Join us tomorrow, June 30th, for the first episode of Sanity on Sanity, where you get to ask questions to Saskia and Peter, the two engineers behind this living intranet. Already in our server? It's under Events — hit Interested for a reminder. New to Discord? Click Join Server, then Events → Interested. Discord event link 👉 https://lnkd.in/eyvye9iH Join our Server 👉 https://lnkd.in/eMWJcjjJ

Lawyers love Word. So how do you earn their trust with AI? At Pioneers London, Timothy Avni (Planes Studio) shared how his team shipped AI into a regulated industry and won client buy-in in two weeks. He walked through: ➡️ Delphius: the agentic legal platform 150+ organizations now rely on, from global banks to big tech. ➡️ AI evals in Sanity: AI runs contract reviews, and lawyers evaluate them and weigh their judgment against the model's. ➡️ Why it still works: Delphius has run for 18 months and counting. Tim's point: once your content is structured properly, the AI on top becomes the easy, swappable part. As Tim says, "what happened next was pretty dope." See what he means on our YouTube 👉 https://lnkd.in/e4suirib

Agent skills get talked about as a developer thing. I think that undersells them. A skill is really just a standard operating procedure written down so an agent can act on it. How you run a launch. How you turn one announcement into eight posts. The company's strategy and the style guide. The stuff that usually lives in someone's head or a doc nobody opens. This stuff can now be pulled in dynamically when relevant into folks AI work streams. The problem is where they live. If skills sit in a Git repo, everyone without a GitHub account is locked out, which is probably most of the company. And it's not exactly easy to add or keep up to date either. So at Sanity we made skills something anyone can publish and the company can actually govern. Write and update a skill in Sanity Studio, hit publish, and it syncs to everyone's Claude across the org automatically. No branch, no PR, no YAML. Now, we can update skills with the Sanity MCP and Content Agent right from Slack. It's pretty cool. How we built it, and how you can do the same: https://lnkd.in/gc_ACif2 Questions? Is this something you'd want to try? Let me know!

Wrote about how we've started collecting product feedback from agents using the Sanity MCP. TL;DR If you give agents the right tool (`give_feedback`) and the ability to explain _why_ they made a tool call (optional `intent` param), you'll learn a lot more than what traditional telemetry tells you. https://lnkd.in/g_5s67aa

"Everyone's a Pioneer." Sanity Pioneers London brought a full room of developers, builders, and operators together to dig into what they're building with AI and Sanity. Thank you to our speakers and game conductors: Timothy Avni (Planes Studio), Steve Ruiz (tldraw), Jarod Reyes, and Peter Cassidy (Piing). And to our community, for the time, curiosity, and expertise you brought into the room. As Kadi Kraman put it: "It helps talking to people who have the same problems. Just great to have the camaraderie." The individual talks land on our YouTube soon. In the meantime, we're enrolling the next cohort of the Pioneers program soon, so if you haven't applied yet, now's a good time. (video featuring Dominic Slim, Lena Pia Hammerstingl, and Kadi Kraman)

Of course we run Sanity on Sanity. That's also the name of our new series, where we show you the architecture behind the applications we build. Episode 1 goes inside Sanity Home, our own intranet. Engineers Saskia Bobinska and Peter Hofstee will walk you through this "living intranet" where information stays current on its own, and the agent finds you what you need. No stale content. No policy buried in a wall of text. See how it runs: scheduled and document-triggered functions, the shared component and data layers the team reuses across every internal app, and the chatbot that took about 15 minutes to wire up with Sanity Context. You won't leave empty-handed. You'll get a GitHub repo of the functions behind Sanity Home with a diagram of how they connect, plus a knowledge base starter built from the same blocks, ready to clone and make your own. Join us live on June 30th in Discord to watch the episode, followed by a Q&A with Peter and Saskia. Already in our server? See link below or head to Events → Interested. New to Discord? Click Join Server (free, ~2 min), then Events → Interested. Discord event link 👉 https://lnkd.in/eyvye9iH Join our Server 👉 https://lnkd.in/eMWJcjjJ