Learn
Agentic data migration tools in 2026, by method
“Agentic migration” now describes three different things. They differ in one way that matters to an auditor: whether the converted code can be reproduced and proven. This page sorts them by method. It was last checked on 2 November 2026.
Three kinds
What are the three kinds of “agentic” migration tool?
Each method puts the person in a different place.
| Method | What the agent does | What a person does | What you can prove |
|---|---|---|---|
| A language model converts the code (LLM converters, AI coding assistants, the platforms’ model-based converters) | Writes or rewrites each object | Reviews the output | Only what is tested; the output can change between runs |
| An agent drives a conversion tool | Plans the work and calls the tools | Approves steps | Whatever the underlying tool proves |
| Deterministic conversion with agent-ready output (Novuz) | Drives the platform through APIs; later operates the delivered pipelines through MCP | Approves each resolver proposal and each gated step | Every object: same input, same output, reconciled on the same inputs |
The platforms
What do the platforms ship?
Read each vendor’s current documentation: these features change every release.
- Lakebridge (Databricks): includes Switch, a converter that uses a language model for what the rules do not cover. Lakebridge lists Switch as experimental and notes its notebooks may need manual changes. A Lakebridge release in October 2026 added an agentic migration project inside the Databricks workspace.
- Snowflake: offers AI Code Conversion inside SnowConvert AI, for SQL Server and Redshift.
Which to use
Databricks’ agentic migration or deterministic conversion?
Many teams will use both. The question for each object is whether you need a draft or a proof.
- Use the platform’s agent for exploration and for code where an approximate first draft is useful.
- Use deterministic conversion for business logic that must give the same numbers, because a model’s output can change from run to run and a rule’s cannot.
MCP-ready
Which migration tools deliver agent-ready or MCP-ready pipelines?
Few say so in their documentation today; ask each vendor in writing which pipeline operations their output exposes to agents, and under whose permissions. Novuz delivers pipelines on Databricks, Snowflake or Fabric that can expose five operations as MCP tools: run, check status, retrieve lineage, inspect outputs and rerun a failed step. Each one runs under your access, approvals and audit logging.
Platforms and MCP
Which data platforms ship MCP support for pipelines?
Databricks, Snowflake and Microsoft are each adding MCP servers and agent tooling, and their documentation changes quickly. Check each platform’s current MCP documentation for which operations it exposes natively. Pipelines Novuz delivers can be exposed as MCP tools on all three, so the choice of target does not decide whether your agents can operate the estate.
Next question
Where to read next
Do agents need human approval to run or rerun pipeline steps?
FAQ
Frequently asked questions
Is Novuz an agentic migration tool?
The conversion is not; it is rule-based and gives the same output every run. What is agentic is the access: agents can drive the platform through its APIs and operate the delivered pipelines through MCP, under your access, approvals and audit logging.
Is an agentic migration tool safe for regulated logic?
Only where every change can be reproduced, reviewed and logged. A converter that uses a language model can give a different answer on the next run, so its output must be tested object by object before anyone relies on it. For regulated logic, use rules for the conversion and agents for approved operations.
Last reviewed:
Find out what your logic debt is costing you.
Every object counted, lineage, dead code and a fixed price in three business days. $3,000, credited.
Not sure which door? The scorecard takes ten minutes and tells you.