Consume AI Builder Models Natively in Canvas Apps with Power Fx
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?
Community Insight
The question tests native AI Builder integration in canvas apps; the trap is assuming you need custom code (PCF) or a cached result instead of the built-in Power Fx function the model automatically exposes.
When a canvas app must call an existing AI Builder model during live user interaction, the native path is to reference the model as a Power Fx prediction function (for example, a 'Predict'-style call bound to a control's OnSelect or a gallery's output). This page explains why invoking the model with Power Fx is the correct answer (C) and why stored predictions, PCF components, and retraining are wrong.
Learners often pick B (add a PCF code component), reasoning that custom code is needed to 'call' the model — but a PCF component is not native consumption and adds unnecessary custom development when the model is already exposed as a Power Fx function.
Community Discussion (1 comments)
Comments & Corrections
No comments yet — spotted an error or have a note? Share it below.
Expert Analysis
Why the Answer Is Correct
Every trained and active AI Builder model is automatically surfaced in canvas apps as a Power Fx function that returns predictions when you pass the required input (here, the image). Writing that function call in a formula—typically on a button's OnSelect or in a label's Text property—makes the app call the model live during user interaction, which is exactly what "consume the AI model natively" means. Option C describes this direct, no-code integration path.Why the Other Options Are Wrong
A. Reusing a previously stored prediction returns cached historical results and does not generate a new prediction at interaction time, violating the requirement. B. A PCF code component is a custom-code extension; it works but is by definition not "native" consumption, and it is unnecessary because the model is already callable from Power Fx. D. AI Builder models are trained once and reused; retraining per submission is unsupported, extremely slow, and unrelated to making predictions in the app.Community Comment Notes
The single community comment agrees with the verdict: the responder answered "C. Invoke the model by using Power Fx." and received a like, matching the documented Microsoft approach where a deployed AI Builder model becomes a Power Fx function callable directly inside the canvas app.Official Reference
Exam Strategy
On exam items, the word "natively" is the deciding clue: it points to built-in platform capability (a Power Fx function) rather than custom code (PCF) or external workarounds. Always confirm the requirement about timing too—"during user interaction" rules out any stored or cached result option.
Frequently Asked Questions
Why is adding a PCF code component wrong if it can call the model?
A PCF component is custom code and therefore not native consumption. The model already exposes a Power Fx function, so custom components add needless complexity.
Can I reuse a stored prediction instead of calling the model again?
No. A stored prediction returns an old result; the requirement is to generate new predictions in real time while the user interacts with the app.