September 2026 Summaries

Sanity’s September updates focused on bringing AI-assisted editorial workflows, project management, and security oversight closer to content operations. At its Everything NYC conference, the company introduced Knowledge Bases in Sanity Context, which extracts and organizes information from content, websites, or PDFs for agents to access through MCP endpoints, and announced the beta of Workflows, a code-defined and GROQ-queryable system for managing editorial stages. Other releases include a Linear integration for Content Agent, enabling it to complete editorial tasks from Linear issues and comments; sanity.new, which lets agents create projects that users can claim within 72 hours; and a security dashboard covering items such as CORS settings, API tokens, access, SAML, and backups. The update also highlighted Studiofront, a free open-source macOS menu bar app that provides quick access to Sanity Studios across organizations and projects, while emphasizing Sanity’s broader goal of making its platform a central workspace for content and trustworthy AI agents.

Sanity announced several products aimed at making AI agents more reliable, governed, and integrated with organizational content systems, following lessons from its own experience with inflexible third-party agent tools. Knowledge Bases, now in beta, consolidates datasets, websites, and files into structured fact outlines that identify contradictions and provide agents with targeted context through MCP connectors. Also in beta, Workflows lets teams define version-controlled, queryable content-review stages that humans and agents share under the same permissions and audit trail, while Content Variants enables audience-specific content versions to be delivered through existing queries without duplicating documents. The company also previewed Durable Functions for resilient long-running automations and expanded Blueprints support for configuring content infrastructure as code. Sanity Labs introduced Resonance, an audience-focused content auditing tool, and Project Horizon, an experimental environment for persistent, delegating agents with long-term memory. The releases reflect Sanity’s broader vision of a “Content Operating System” in which content, people, and agents operate through common processes, controls, and records.

Sanity announced a suite of products aimed at making AI agents more reliable, governed, and useful in content operations, following its Everything *[NYC] event in Brooklyn. Its beta Knowledge Bases feature organizes datasets, websites, and files into structured fact maps, flags contradictions, and gives agents domain-specific context without repeated searches, while Workflows adds version-controlled, queryable approval stages that humans and agents can follow through shared permissions and audit trails. Content Variants, also in beta for selected customers, enables audience-specific content versions within a single document and resolves them at query time. Upcoming Durable Functions will support fault-tolerant, long-running automations, and expanded Blueprints will define broader content infrastructure as code. Sanity Labs also introduced Resonance, an audience-focused content auditing tool, and Project Horizon, an experimental framework for long-lived agents with memory and delegation capabilities. The company reports rapidly growing MCP usage, with agents making 1.2 million Sanity tool calls in August, and positions these releases as components of a unified “Content Operating System” where automation handles repetitive work while people retain control over decisions.

As organizations deploy AI agents for business-critical tasks, the central challenge shifts from delivering context through RAG pipelines or repositories to governing the content that becomes agent context. Drawing on Andrej Karpathy’s concept of LLM-maintained knowledge bases, the piece argues that scalable systems require controlled ingestion, accessible interfaces, reliable question-answering, traceable outputs, consistency checks, and supporting infrastructure. Organizational information is vulnerable to entropy and drift when facts are duplicated, outdated, contradictory, or buried in informal channels, while AI-generated synthesis can further obscure sources and decisions. The author recommends treating agent knowledge management primarily as a ContentOps responsibility, with clear ownership, approvals, provenance, audience-specific access, and processes for retaining human corrections across rebuilds. Although repositories and DevOps tools can work for small, technically managed setups, extending them to nontechnical teams often recreates a content management system within Git, making intentional content operations essential for maintaining accurate, auditable, and safe agent behavior.