How to Natively Consume an AI Builder Model in a Canvas App
Short answer
D Call the existing AI Builder model from the canvas app's Power Fx logic so predictions are generated live during user interaction.
Question
A company builds a canvas app that processes images submitted by users and must classify them using an existing AI Builder model. The app must generate predictions during user interaction. You need to consume the AI model natively. What should you do?
Key Takeaways
The question tests native AI Builder consumption in canvas apps: the trap is confusing live, on-demand prediction with custom code components or reusing cached results.
AI Builder models are consumed natively inside canvas apps by invoking the model from app logic (Power Fx / the model's prediction control) so predictions are generated live as users interact. This page explains why calling the model from app logic is the correct approach and why PCF components, model refinement, and stored predictions are not.
Common mistake: Choosing A (PCF code component) is the most tempting wrong answer, because a PCF component can display AI output; but PCF is custom code, not native AI Builder integration, and the requirement explicitly says the app must consume the model natively.
Expert Analysis
Why the Answer Is Correct
Option D is correct because a canvas app that must "consume the AI model natively" and "generate predictions during user interaction" should call the AI Builder model directly from app logic. When an AI Builder model is added to a canvas app, Microsoft exposes it as a callable function/control that Power Fx formulas invoke on demand—for example in the OnVisible, OnSelect, or OnChange property—returning live predictions (labels, confidence scores, extracted values) for the image the user just submitted. This is the documented native path: the model is referenced from the app's own logic, no external code is required, and predictions are computed at interaction time rather than read from storage.Why the Other Options Are Wrong
Option A (Add a PCF code component) achieves functionality through custom TypeScript/React code, which by definition is not "native" consumption; PCF is used when you need bespoke UI, not to invoke an AI Builder model. Option B (Refine the model with additional images) improves model quality over time but does nothing to consume the model at runtime or generate predictions during user interaction. Option C (Reuse a previously stored prediction) returns historical data cached from an earlier run, which directly contradicts the requirement to generate predictions during user interaction for the newly submitted image.Community Comment Notes
The single community comment agrees with this analysis, answering "D. Call the model from app logic." and treating direct invocation from canvas app formulas as the native AI Builder consumption pattern. No dissenting comments were posted, so the vote data neither supports nor contradicts the technical reasoning, which stands on the documented canvas-app model consumption behavior.Official Reference
Exam Strategy
Key on the phrases "natively" and "during user interaction": native means the built-in AI Builder model integration (invoked via Power Fx), and live interaction means the model runs now, not from cache. Eliminate any option describing custom code (PCF), offline model improvement (refinement), or historical data (stored predictions).
Frequently Asked Questions
Why is adding a PCF component not considered native AI Builder consumption?
A PCF component is custom code you author yourself. Native consumption means using the built-in AI Builder integration that Power Fx calls directly, with no custom component required.
Why can't I reuse a previously stored prediction in the canvas app?
Stored predictions are results from an earlier run on different data. The requirement is to generate predictions during user interaction, so the model must be invoked live for the current image.
Discussion 1 contributions
Compiled from public community threads and reader submissions, lightly edited for clarity. Handles and vote tallies are shown as recorded.
From public community threads 1
-
exam_upskill 👍 1 Answered DD. Call the model from app logic.
Reader comments & corrections
No reader corrections yet — spotted an error or have a note? Share it below.