We're hiring a senior engineer for Agent Foundations, the team responsible for how agents work with structured content at Sanity. We own two products. Content Agent runs content operations (bulk edits, audits, translation, publishing) through conversation in Sanity Dashboard, Slack, Linear and a headless API. Sanity Context (including Knowledge Bases) is an MCP endpoint that gives any agent structured retrieval and semantic search over a team's live content, whether it be Sanity Documents or your own federated content from outside Sanity. We're looking for someone who's built AI systems people can depend on in production.

At Sanity.io, we build an AI-powered content operating system that lets teams model, create, and automate content the way their business actually runs. SKIMS, Figma, Riot Games, Anthropic, COMPLEX, Nordstrom, and Morningbrew run their content operations on it.

Most AI products add a model to a content store. We work on what sits underneath. We build knowledge bases from Sanity datasets and external sources, and retrieval that combines GROQ filters with keyword and embedding search. Our agents need to understand schemas, references and validation rules. Telemetry and evals tell us how well the systems work. You'll work across these systems, owning your domain end to end and working closely with product and design to decide what gets built next.

What you'd be doing

Build retrieval and knowledge bases. Turn sources into documents and chunks. Develop how knowledge bases are created, kept fresh and scoped, and how agents query them using GROQ, keyword and semantic search together.

MCP endpoints and tools give agents their interface to our systems. You'll build these, along with tools and integrations that extend Content Agent.

Work on agent loops. Write the orchestration logic that controls an agent's actions and their order, including when to bring a human in.

Engineer system prompts. Give agent instructions the same engineering attention as the code around them.

Measure and improve reliability, retrieval quality and cost through evals, telemetry and experiments. We measure it in production as well as on test sets.

Ship across the stack. Build interfaces in Dashboard and Studio, APIs and backend services. Work on Blueprints, the infrastructure-as-code path customers use to define their agents and knowledge.

Help shape the products. Sanity Context is new and changing quickly. You'll have considerable latitude to decide what it and Content Agent become.

About you

Location: London, UK or Oslo, Norway (in-office 2–3 days/week)

You're strong in TypeScript and React and comfortable working across frontend, APIs and backend (often in Go).

You've shipped LLM-based systems in production and understand the work it takes to make them behave reliably for every user.

You understand chunking, embeddings, hybrid search and ranking, including where RAG falls short. Experience with search or information retrieval is a strong plus.

You think through a capability's user intent, edge cases, failure modes and cost before deciding on the architecture. That includes what the agent needs to know, which tools it can use and when it should escalate.

You write well and pay attention to how phrasing changes an agent's behavior. You can write precise instructions for a system that interprets language literally.

You're comfortable working with probabilistic systems. You design evals, run experiments and interpret results statistically, using aggregate trends to decide which failures need attention.

You have experience with MCP, tool-use patterns and agent architectures, or you're curious enough to explore them.

You care about the quality of the code and products you ship, down to the writing.

You don't need to meet every requirement to apply. We're especially interested in people who've developed engineering skills after starting in fields such as writing, interaction design or liberal arts. Some of the best agent work we've seen comes from people who understand language and how people work, alongside systems architecture.