Last week, Dataverse received a new preview feature that has been sorely missed over the years when configuring security models for CRM customers. It’s the ability to handle exceptions, not just hierarchies and groups based on record ownership. You know, “show users all the records, except these ones tagged as confidential” kind of business requirements. Some systems might call it row-level security (RLS). In short, Dataverse now lets you define permissions using filters on record fields, rather than relying on record ownership.

This sounds huge, and it is! When I started looking deeper into the feature, though, it triggered a thousand little papercuts. It’s not just about one feature but rather the state of things in the low-code platform UX that has been left to rot for too long. Today, with the all-code approach becoming ever more popular, it raises the question of whether the Maker portals help in adopting such new features or if they have become a hindrance instead.

Sometime toward the end of 2024, this feature got added to the wave 2 plans: “Secure data access using filtered views”. Hey, that sounds pretty interesting! Why didn’t I see anyone talk about it? So, I did what I do: took a screenshot and wrote a LinkedIn post about it.

MVPs also started covering the feature. Then, in July 2025, when the feature still had not landed in preview, I went hunting for it. And it was gone. Deleted without a trace, not even an entry left in the change history page of MS Learn release plans. Usually there was at least a “deprioritized and will not be delivered” entry. This time there was nothing. So, I had to create the entry myself:

My read was that the product team was struggling to get any funding for features that didn’t connect to Copilot or agents. Even for functionality that had been requested by customers over and over again, there seemed to be no development budget available for building platform functionality that you couldn’t associate with AI chatbot scenarios (which most customers weren’t exactly asking for).

Over a year went by. During that time, release plans and release waves were deprecated, replaced with a promise of an “always-on” AI at Work roadmap. After 2026 release wave 1 was done (end of September), there weren’t going to be any new release trains carrying fresh new release plan item cargo.

Yet somehow, out of the blue, the filtered record ownership (preview) documentation page appeared at the end of September:

Don’t bother looking it up on the AI at Work roadmap. There’s one single feature listed under Power Platform governance and administration and it’s not this one. (It’s about Copilot Studio, of course.) Just be happy that the feature finally landed.

The point of this newsletter issue isn’t the lack of transparency or inaccuracy in roadmap comms, though. I’ve covered that many times over the years. What I wanted to write about instead is what these new platform features in Dataverse are like in 2026 and what that tells us about the dilemma low-code is facing.