← back
From Systems of Record to Systems of Context — Omri Bruchim & Tomer Ast, monday.com
Takeaway
Useful workplace assistants need continuously maintained relationships and priorities, beyond access to raw records.
Summary
- monday.com argues that retrieving tasks, messages, and meetings is insufficient for an assistant to determine a user's priorities.
- Its world model represents work entities and dependencies, current urgency signals, and a durable profile distilled from decisions and work patterns.
- A slow engine analyzes weeks of activity to learn roles, routines, collaborators, and goals, while a fast engine frequently refreshes short-window signals.
- Precomputing these relationships gives Monday Sidekick context for reasoning before a user asks what to focus on.
context-engineeringenterprise-agentsmonday-com
Original description
Ask your AI assistant what you should focus on right now and you get a list of disconnected bullets dressed up as a confident paragraph. When Omri Bruchim tried it, Claude told him to go to the gym. The assistant has every board, task, email, and Slack message you have ever touched, and still zero understanding, because the problem was never retrieval. It is that a system of record stores what happened but not what it means. monday.com's answer is to become a system of context. They build that context layer ahead of time from two engines. A slow engine mines weeks of activity into a durable profile of who you are and how you work; a fast engine reads the last few days for what is suddenly urgent and who you are pulled in with. One knows you, the other knows your day, a split that shows up in neuroscience as the hippocampus and neocortex and in data systems as lambda architecture. The context is precomputed and served to their agent, so it degrades gracefully, judges when to speak up, and compounds as every new day and source sharpens the model. Speaker info: https://x.com/omribruchim / omribruchim https://edginary.io / tomer-ast Timestamps: 0:00 - From system of record to system of context 0:56 - The gym answer: data without understanding 2:36 - monday.com, Sidekick, and where work lives 4:31 - Three reasons context is hard 7:19 - The Monday world model 8:15 - The data model and its two engines 10:21 - Why the split mirrors the brain and lambda architecture 11:14 - How it comes together, and the honest limits 13:22 - Answering the question with Sidekick