Krux
March 15, 2026
Notion Engineers Now Get Answers at 2 a.m. Without Pinging Anyone
Published: March 15, 2026 at 12:32 AM
Updated: March 15, 2026 at 12:32 AM
100-word summary
Notion's internal case study reveals how one person building a Custom Agent spiraled into "tons of agents running across the company." Engineers now get answers at 2 a.m. without waiting. Customer support fields repeat questions with instant replies pulled from knowledge bases. The trick? Start small. Notion teams began by offloading a single annoying coordination task, testing it as a basic prompt before converting it into an agent. Once proven, they turned it into a template and let champions spread it. Custom Agents watch Slack for triggers (new messages, emoji reactions, thread starts) and run recurring workflows on any cadence. The internal adoption pattern suggests AI assistance spreads fastest when...
What happened
Notion's internal case study reveals how one person building a Custom Agent spiraled into "tons of agents running across the company." Engineers now get answers at 2 a.m. without waiting. Customer support fields repeat questions with instant replies pulled from knowledge bases. The trick? Start small. Notion teams began by offloading a single annoying coordination task, testing it as a basic prompt before converting it into an agent. Once proven, they turned it into a template and let champions spread it.
Why it matters
Custom Agents watch Slack for triggers (new messages, emoji reactions, thread starts) and run recurring workflows on any cadence. The internal adoption pattern suggests AI assistance spreads fastest when it solves one specific pain point first, not when rolled out company-wide on day one.