Serve live, governed data in AI-built apps with Amazon Quick
Amazon Web Services Wei Kuo
Amazon Quick apps can now query governed Quick Sight datasets live, not from frozen snapshots. That means current numbers and row-level access rules finally travel with the app.
Based on reporting by Amazon Web Services, Wei Kuo — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AWS is adding live governed data to Amazon Quick apps, and that fixes the biggest flaw in the earlier version: published apps were stuck showing whatever numbers were baked in when the builder hit publish. Connectors, web search, AI inference, and content sources already ran at view time. Now Quick Sight datasets can do the same.
The pitch is simple. A user describes an app in plain language, the agent builds it, and the finished app re-runs the SQL each time someone opens it. The app can pull from governed Quick Sight datasets in SPICE or Direct Query, and it does the query as the person viewing it. That matters because row-level and column-level security stay in force per user, instead of being flattened into one shared snapshot.
AWS uses a sales-renewals example to show why this is useful. A regional sales leader can ask for a live app that lists customer renewals in the current quarter and the next six months, then drill into revenue and margin from SaaS Sales data. The agent discovers the datasets, writes the SQL, asks for consent by dataset name, and then builds the app. The same workflow can be extended with enterprise content and AI inference, such as overlaying a strategy document to help judge whether a deal should move forward or get a discount.
The admin side is fairly strict. Viewers must be authenticated Quick users, anonymous access is out, and each viewer gives one-time consent per dataset on first use. Existing RLS and CLS rules apply automatically, consent is checked server-side on every query, and the app won’t build if the builder’s RLS returns no data. If a query grows too large, Quick will say to narrow it rather than hand back a chopped-up answer. Direct Query datasets from different sources also can’t be mixed in the same app.
This is less about flashy AI and more about finally making the app honest. A live app that respects the viewer’s permissions is a much better default than a pretty screenshot with stale numbers. And yes, that still leaves the boring work of consent, guardrails, and dataset hygiene. Which is exactly where the boring work belongs.
My take — AI-written commentary, not fact-checked reporting
This is the right move: an AI-built app is only useful if the numbers move with reality and the permissions do too. Frozen snapshots were fine for demos and terrible for decision-making, which is a very enterprise way to learn a lesson the hard way. AWS is basically saying the app can be smart, but it still has to ask nicely.
Read more about this at: Amazon Web Services