In defense of agent sprawl

Sanity describes how an internal hackathon produced roughly 80 AI agents across departments, arguing that widespread agent creation is manageable when agents are lightweight, replaceable interfaces rather than isolated stores of knowledge. Its @s platform routes Slack requests to specialized agents that automate tasks such as investigating outages, assessing support incidents, provisioning demos, answering policy questions, connecting product goals to customer feedback, and locating internal expertise or terminology. The company attributes the initiative’s success primarily to shared, team-owned Knowledge Bases that preserve sourced and maintained facts independently of any individual agent, preventing conflicting or lost information when agents are retired. It uses Studio-based validation, tool reviews, restricted access, zero-retention model calls, and usage monitoring as safeguards, while recognizing that connectivity to data sources and locating authoritative context are greater obstacles than writing prompts. Sanity reports that most agents functioned at least partly on their first day, but emphasizes that reliable, canonical organizational knowledge, clear ownership, transparency about knowledge gaps, and human review are essential—especially as agents gain autonomy. The company recommends centralizing facts while keeping models, frameworks, and agent implementations interchangeable through MCP-based Knowledge Bases.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.