# Automating Team Knowledge Management with AI Agents Reducing workplace interruptions by deploying a custom Slack agent for internal queries. - **Source:** [UX Collective - Medium](https://brifing.si/source/5c6bdc88-e150-4203-af43-069801e37dd9) - **Published:** 2026-11-05 - **Topic:** [technology](https://brifing.si/c/technology) - **Original article:** - **On Brifing:** https://brifing.si/s/c42cf4f4-035f-4f7a-88d2-8385d0ac9a71 ## The Burden of Expertise Every team relies on a central expert who manages institutional knowledge and answers repetitive questions. For a remote design team lead, these constant interruptions caused significant frustration and reduced productivity. Research indicates that while workers may speed up to compensate for frequent breaks, they experience higher stress and mental fatigue. ## Analyzing Internal Queries A review of 2,413 requests over two and a half years revealed that 53% of questions were redundant or easily answerable via existing documentation. Another 35% involved routine legwork, while only 12% required the lead's specific judgment. This data highlighted a clear opportunity to automate common housekeeping tasks using AI. ## Building the Prototype The team developed a Slack agent named 'le goat' to intercept routine questions. Using a Cloudflare server and Gemini, the bot queried Notion, GitHub, and web resources to provide answers. Initial tests showed the bot could retrieve information, but it often generated overly long, essay-like responses that required further refinement. ## Refining Interaction Design To improve usability, the team applied principles from Google’s People + AI Guidebook. They implemented features like a 'Thinking' status indicator, concise one-line answers, and clear source citations. The bot was also restricted from making autonomous decisions, instead drafting updates for human approval to ensure accuracy and maintain team control. ## Pilot Program Results During a two-week pilot, the team asked 228 questions, averaging 31 per day. The agent was eventually integrated into Figma, allowing designers to query documentation directly within their primary workspace. This shift moved the agent from a simple Slack bot to a tool embedded directly into the team's creative workflow. ## Addressing AI Uncertainty Testing revealed that while the agent did not hallucinate confidently, it frequently reported uncertainty. Analysis showed this stemmed from conflicting data across sources, such as outdated PRDs versus current code. This exposed a deeper coordination problem where documentation failed to keep pace with the actual evolution of project specifications. ## Proactive Knowledge Management The agent evolved into a proactive teammate that shares daily updates on changes and flags discrepancies between sources. By identifying conflicting information, the bot helps the team maintain alignment. It now automates the identification of stale documentation, allowing human owners to review and fix errors without manually auditing every single file. ## Technical Architecture Details The system uses a small, fast model called Jev to classify incoming messages before routing complex queries to a larger LLM. Connectors were expanded to include Supabase, Notion, and GitHub. Privacy controls were implemented to allow users to opt-in to searching private conversations while keeping sensitive data restricted to authorized users. ## Measuring Operational Impact Post-implementation metrics showed a decrease in direct questions reaching the lead, reduced wait times for team members, and improved accuracy in responses. By moving answers into shared threads, the team created a searchable knowledge base. This transition effectively reduced the cognitive load on the team's most experienced members. ## Scaling the Agent Farm The success of the initial agent has led to plans for specialized bots, including a critique-focused owl and a task-oriented mule. This 'agent farm' approach aims to distribute different types of support across the team. Each agent is designed to handle specific operational roles, further freeing humans for creative work. ## Strategic Value of Automation Deploying AI agents for internal knowledge management is not just about saving time; it forces teams to maintain better documentation. As agents become integral to daily operations, the quality of the underlying knowledge base becomes critical. Reliable data ensures that both humans and AI can function effectively within the organization. ## Sources covering this story - https://uxdesign.cc/everyones-got-a-quick-question-374a736ff7cb?source=rss----138adf9c44c---4 — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._