Descriptions:
Engineering managers Omri Bruchim and Tor from monday.com presented at AI Engineer on how the company is transforming its work management platform from a system of record into what they call a system of context. The core provocation: monday.com already has access to users’ tasks, Slack messages, meeting notes, project boards, and email — yet AI assistants built on top of that data still can’t answer a simple question like ‘what should I focus on today?’ The problem, they argue, isn’t missing data or inadequate retrieval. It’s the absence of understanding — knowing what the data means and how pieces relate to one another.
To solve this, monday.com built the Monday World Model, the intelligence layer powering Monday Sidekick, their AI personal assistant. The architecture uses two engines operating on different time horizons. A slow engine analyzes weeks of user activity to build a durable profile: the user’s work patterns, collaboration graph, goals, decision-making cadences, and role-specific priorities. A fast engine reads real-time signals — what’s overdue, what’s urgent right now, which collaborators are actively engaged — to understand what’s happening today specifically.
A key architectural insight is that this understanding must be built proactively, well before a user asks a question. Attempting to construct contextual understanding at query time is too slow and too shallow. The talk also draws a sharp distinction between retrieval (fetching data) and understanding (knowing what it means), arguing that the AI industry routinely conflates the two — and that conflation is why even data-rich assistants still give generic, disconnected answers.
📺 Source: AI Engineer · Published July 22, 2026
🏷️ Format: Deep Dive
