Power Apps Vibe

Status: Public Preview — retiring. Apps stay accessible until 30 November 2026.

What it does

An AI-native authoring surface at vibe.powerapps.com where you describe a business problem in natural language and it produces a plan, a data model and a working app with generated code. Plan, data and app stay linked, so a change to one updates the others. When you want out of the generated experience, you can download the source and convert the app into a Power Apps code app and carry on in your own IDE.

Key facts

Retiring (October 2026). Microsoft's migration guide says Power Apps Vibe at vibe.powerapps.com is retiring and that you can access your vibe apps until 30 November 2026. The overview page still reads as a preview with only the September pointer to Cowork, so the migration page is the one to trust. Migration is a rebuild, not a move: you download the source zip and attach it to a migration prompt in Copilot Cowork (Frontier program only) or Apps in Copilot Studio (public preview). Data in Dataverse tables or SharePoint lists stays put and the new app connects to it. Draft tables, data connections and ALM support are not carried over, so there is no route to a test or production environment yet. An automated migration option was planned for mid-October.
Microsoft is steering people off it (September 2026). The overview page now carries an IMPORTANT note saying app building is available in Copilot Cowork and will roll out to Copilot Studio soon, and that those are the recommended experiences going forward. Vibe hasn't been retired, but it is no longer the direction of travel.
Reached through vibe.powerapps.com, or from the maker portal via Try new experience (Preview) — which you have to re-enable at every sign-in if you go that route.
A tenant admin must switch on Copilot in Power Apps (preview) in Power Platform admin centre tenant settings first.
Not available in the default environment. At preview it runs only in US, Australia, Asia and India regions, and only in English.
Microsoft recommends enabling External Models (Anthropic models) per environment for better output quality — Environments > Settings > Features > Enable External Models.
Three modes matter in the prompt box: straight generation, Plan mode (review and discuss before anything changes), and Enhance prompt to expand a thin description before generating.
Download source in the app editor gives you a zip. From there pa app init, pa app run, npm run build, pa app push turns it into a code app.
Converting creates a new, disconnected app. Changes in the code app never flow back to the vibe app or its plan. It's a one-way door.
Vibe-generated apps may reference images on res-dev.cdn.officeppe.net, res-sdf.cdn.office.net and cdn.hubblecontent.osi.office.net. Those hosts must be allowlisted in the code app Content Security Policy in the target environment or images silently disappear after conversion.
Conversion prerequisites are developer tooling, not maker tooling: Node.js LTS, the Power Apps CLI, an IDE, and an environment with code apps enabled.

When to use / skip

Don't start anything new here. With a retirement date on the calendar, this is now a migration topic rather than a selection one.

Use it for the front half of a build — getting from a vague requirement to something a stakeholder can click, fast, with the data model roughed in. It's genuinely good at collapsing the discovery-to-prototype gap, and Plan mode makes it usable in a workshop rather than just a demo.

Skip it as a delivery mechanism today. Preview, region-limited, English-only, no default environment, and a one-way conversion. If the app matters, treat vibe as the sketch and the code app as the deliverable — and budget for the conversion, because it isn't a button.

Configuration decisions

Whether to enable external models. It's an environment-level switch that sends work to Anthropic models. Better output, and a conversation your client's security function will want to have first, not after.
Which environment. Not the default one, and it has to be in a supported region — which for a lot of European clients rules the whole thing out at preview. Establish this before you scope anything.
Where the app is meant to end up. Decide at the start whether this is a throwaway prototype or the seed of a code app. The answer changes how much effort is worth spending inside vibe, because none of that polish survives cleanly.
Who converts, and when. The conversion needs a developer. If nobody on the account can run a CLI, vibe's output has no exit.
CSP allowlisting in the target environment, if the generated app uses stock imagery.

Gotchas

Deprecated 30 November 2026: access to Power Apps Vibe ends. Publish any draft tables before you leave, because unpublished ones don't migrate, and keep the original app until the rebuilt one is confirmed. See migration guide.
Rebuilt apps can't yet connect to Dataverse tables in the default environment (the Microsoft page carries an unresolved engineering note on this), so check where the tables live before promising a like-for-like rebuild.
The rebuild prompt is a spec, not a copy job. Vague wording such as "similar to" invites a redesign. Name the tables by logical name and attach the zip unextracted.
Don't start a client on vibe now without saying out loud that Microsoft has pointed at Copilot Cowork and Copilot Studio instead. A preview the vendor has already recommended moving away from is a different risk conversation from a preview that's simply early.
The disconnection at conversion is the one that hurts. Teams convert early, keep iterating in vibe because it's faster, then discover the two apps have diverged with no merge path.
power.config.json has to be salvaged by hand — copy connectionReferences and databaseReferences out, delete the file, run pa app init, paste them back into the new one. Miss it and the converted app has no data.
Missing images after conversion look like a broken build. They're a CSP problem, and nothing in the error output says so.
Re-enabling the preview experience on every sign-in through the maker portal makes people think it's been withdrawn. Use the direct URL.
English-only is not a soft limitation. It's the prompting language, and it's the reason non-English-speaking teams get poor results and blame the model.

Consultant notes

Inventory every vibe app in the tenant now and put the November date against each owner. Prototypes that quietly became team tools are the ones that get missed.
Rebuild in a new, separate app and run the two side by side briefly. Cowork shares by link to anyone with a Microsoft 365 Copilot licence in the organisation, which is looser than most clients expect, so check the sharing model before you move users across.
Anything that reached production through the vibe-to-code-app conversion is unaffected by the retirement of the authoring site, but check it isn't still being edited in vibe.
The September 2026 pointer to Copilot Cowork changes the recommendation. If a client is choosing today rather than already mid-flight, look at the Cowork app skill first and treat vibe as the fallback for what Cowork can't do yet — not the default.
Sell it as an accelerator for the requirements phase, never as a build tool. The demo is spectacular and clients will hear "we don't need developers"; the conversion path proves the opposite, so show that in the same session.
The region restriction is the first question to ask, before anyone gets excited. A UK or EU tenant with no US/APAC environment cannot use this today, and finding that out after a scoping workshop is an avoidable embarrassment.
Post-go-live risk sits entirely in the one-way conversion. Anything that reaches production must be the code app, with the vibe app formally retired — otherwise you end up supporting a shadow copy that a business user keeps editing.
Sequence it early and hard-stop it. Vibe for week one, convert, then normal code-app ALM. Teams that let vibe run alongside development spend the project reconciling two versions of the same app.

Retirement has a date now — revisit after 30 November 2026 and retire this doc once vibe.powerapps.com is gone.