haven vs Omni

Omni publishes a dated weekly digest whose body is a single line listing that week's items, so each entry compresses several releases into a sentence. Across the window the pattern is unmistakable: AI-powered semantic model generation reaching general availability, AI Routines creatable from chat and deliverable to Slack, AI model suggestion endpoints, AI credit controls scoped to embed entity groups and individual users, AI Evals on Azure, and MCP surfaces appearing both in-app and as a searchDashboards tool. The most recent week breaks that streak — default filters on composite topics, stopping a running dashboard query, full-screen preview editing — the first digest in two months led by conventional BI work.

Two things have been happening in parallel and they are related. Omni pushed AI into the modelling layer rather than only the query layer, which is what semantic model generation reaching GA signified, then built the commercial and access controls those features require — credit limits per user and per embed entity group arrived within weeks of the capabilities that consume them. The MCP work points at a third direction, exposing Omni's content to external agents rather than only serving its own chat. The latest week's return to filters and query controls suggests the AI surface has reached the point where the surrounding product has to catch up to it.

With searchDashboards already shipped as an MCP tool, more of Omni's catalog is the obvious next thing to expose that way, and credit controls should keep extending to cover newer AI surfaces. Whether the non-AI week is a pause or a genuine rebalancing is not something one digest can settle.