A developer reviews an AI-assisted tool converting Oracle SQL code to PostgreSQL on a large monitor.
Microsoft has put AI-assisted Oracle-to-PostgreSQL application code migration into public preview, and it is more than a button in an editor. It is a Copilot-driven workflow that rewrites the code around your database. It does not move your data or convert your schema. Azure Updates lists it as "In preview", which that page defines as available to all Azure customers for non-production use and testing. Treat it as a way to speed up a rewrite, not as a switch you flip.

A developer reviews an AI-assisted tool converting Oracle SQL code to PostgreSQL on a large monitor. What the preview actually does​

The feature targets the Oracle-specific code that sits around a database. That means embedded SQL, connection logic and database-interaction patterns. Microsoft's documentation says it is meant to cut manual rewrite effort in Oracle-to-PostgreSQL moves. It converts Oracle-specific database interaction code into PostgreSQL-compatible code so existing Oracle-based applications can work with Azure Database for PostgreSQL.

Microsoft also warns that completeness varies with application architecture, database usage patterns and Oracle-specific feature usage. The Azure Updates listing itself gives no success rates, pricing or production-availability promise, so none should be assumed.

The workflow has changed shape​

Microsoft's documentation now describes two routes.

  • The newer skill-based route. Microsoft says the next version of the capability ships as a skill in the Postgres Skills marketplace. It works with GitHub Copilot in both VS Code and Copilot CLI. Microsoft recommends it for new projects.
  • The extension route. This is the PostgreSQL extension for VS Code, which opens a Migration panel and hands the work to Copilot agent mode. Microsoft says it remains supported.

Microsoft's migration guidance also describes the skill in terms of two modes, assessment and conversion. A separate Microsoft Tech Community post on the complete migration journey says the application-code step uses the new migrate-oracle-app skill in the postgres-skills plugin. The same post says Java and .NET teams already using Modernize CLI can run the Oracle-to-PostgreSQL application migration there too, now in Public Preview.

Some earlier descriptions of this feature mention specific VS Code and model-version requirements. Microsoft's current pages have moved on, so check the live documentation for the model guidance that applies to you.

Prerequisites for the skill-based route​

From Microsoft's overview and tutorial, the skill-based route expects the following.

  • The postgres-skills plugin installed from the Postgres Skills marketplace. Adding the marketplace alone does not install the plugin.
  • The Microsoft PostgreSQL extension with a saved connection to your destination database. The Copilot CLI route uses its own saved postgres-mcp profile instead.
  • A local Git repository with at least one commit, an attached branch and a clean working tree.
  • Node.js 22 or later, with node, npm and npx on your path.
  • Docker running, for disposable PostgreSQL integration tests.
  • A destination database that already holds the converted schema, plus any schema-conversion artifacts and coding notes you have.
  • A code-capable Copilot model. Microsoft's current pages name a recommended model and reasoning configuration, "where available".

Schema artifacts and coding notes are optional but recommended. They give the agent source definitions, type mappings and routine signatures to work from.

Assess first, then convert​

The tutorial separates the work into stages, and the separation is the useful part.

  1. Open the repository and verify the skill. In VS Code, confirm the skill appears in Chat: Configure Skills and that your saved database connection is available.
  2. Run an assessment. This is database-read-only and does not change source files by default. It looks at Oracle dependencies, architecture, relevant destination objects, and build and test readiness.
  3. Review the findings. Check for missing or incompatible destination objects, failing baselines and missing dependencies. A failed test or missing dependency is not a successful migration check.
  4. Run the conversion. The skill presents the impact and asks for approval before editing source files. It then converts the code in place and uses build and test results to validate the changes.
  5. Read the report and diffs. Reports and file diffs are written into the repository. The skill leaves changes in your working tree and does not stage, commit or push them.
  6. Validate in a non-production environment. Do this against the intended PostgreSQL schema, as described below.

A one-pass assess-and-convert mode exists. Microsoft notes it authorizes reversible source changes without a separate review pause, so use the staged flow if you want to read the findings first.

Opt-in actions and safety boundaries​

Microsoft draws careful lines around what the agent may do to your database.

  • Persistent object creation and target-write testing are separate opt-in actions. Microsoft says they need a write-enabled saved profile and extra safety checks, and that they are off in its examples.
  • Tests are controlled through an options setting. It enables baseline setup, unit and integration test generation, and broken-test repair, but not database writes.
  • A saved connection does not grant extra privileges. The skill does not grant database privileges either.
  • Do not paste passwords or connection strings into chat. Use saved connection profiles.
  • Review your organization's policy on sharing code and database context with AI services. That matters for regulated codebases.

What to review before trusting the output​

Microsoft's best-practices guidance is blunt about the limits. It says completeness varies by application. The tutorial's review checklist names the areas where Oracle-to-PostgreSQL rewrites tend to go wrong:

  • SQL semantics, null handling, type mappings, transactions and generated identifiers.
  • Stored routine calls and any inferred or reconstructed database behavior.
  • Generated tests and their assertions. Microsoft cautions against weakening valid assertions just to get a green result.
  • Changes to build files and application configuration.

Microsoft also says the skill can attempt best-effort reconstruction of routines when authoritative definitions are unavailable. A reconstruction that builds or passes synthetic tests is not proof of equivalence with the original Oracle routine. Docker passes validate behavior against a disposable PostgreSQL instance, not your real destination.

The guidance gives a rough sense of where effort concentrates. Applications that lean on specialized Oracle features, such as Advanced Queuing, Spatial, Text Search, PL/SQL packages carrying business logic, or TAF, FCF and VPD behavior, need architecture decisions as well as syntax changes. Microsoft's advice there is not to assume a syntactic replacement preserves the original feature.

For .NET and native-client codebases, the guidance suggests checking Oracle provider usage against the Npgsql version for your target framework. It also suggests validating Entity Framework configuration and inventorying OCI and ODBC usage.

Where this fits in a larger migration​

Application conversion is one part of a three-part journey. Microsoft's Tech Community post describes converting the schema with the PostgreSQL extension for VS Code, moving data with Azure Database Migration Service, and updating application code with the new skill. Microsoft's tutorial states that application conversion does not migrate production data. Plan data migration separately.

This lands on the same day as another Azure Updates item. WindowsForum has covered Azure DMS gaining an Oracle-to-PostgreSQL route in public preview, which that coverage describes as offline only, so downtime is required. Taken together, the pieces of an Oracle exit now exist, but all are previews.

Practical take​

The sensible reading is that this is a productivity tool for engineers who already own the migration. Its strengths are the Git-native approach, the assessment step and the explicit opt-in boundaries. Its weak spot is the one every AI-assisted rewrite shares: a green build says little about whether business logic behaves the same.

If you want to try it:

  • Start with a non-critical application and a scratch PostgreSQL destination.
  • Convert the schema first and keep the coding notes.
  • Run assessment only, and read it before authorizing edits.
  • Treat every diff as a pull request from a fast but overconfident junior colleague.
  • Compare real query results with Oracle's before anything near production.

Microsoft's own summary applies to every workflow here: AI-assisted conversion can produce incorrect or incomplete changes, and it needs independent review and functional, integration and performance testing before deployment.

 

References

  1. (In preview) Public Preview: AI Assisted Oracle-to-PostgreSQL Application Code Migration Azure Updates 2026-11-17T20:00:32Z
  2. Complete Oracle to PostgreSQL Migration: Move Your Schema, Data, and Application Code techcommunity.microsoft.com
  3. What Is Oracle to PostgreSQL Application Conversion? - Azure Database for PostgreSQL learn.microsoft.com