Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseAPI / Product · ActionableWindsurf Blog→ original

AI Engineer Devin Migrates a 142-Object Platform to Databricks in 8 Days

Moving off a legacy data warehouse (Teradata, Oracle, Snowflake) to Databricks usually eats up an engineering team for quarters. Cognition and Databricks have launched Devin: an AI engineer that rewrites code, tests it, and reconciles data. A Fortune 500 energy company moved 142 objects into Unity Catalog in 8 days. If a migration is stuck in your backlog, it is worth piloting on a single domain.

AI-processed from Windsurf Blog; edited by Hamidun News
AI Engineer Devin Migrates a 142-Object Platform to Databricks in 8 Days
Source: Windsurf Blog. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceWindsurf Blog ↗
Editor-in-Chief
Key metric #1Devin runs hundreds of tasks in parallel and tests code in the customer's environment
Pre-launch reconciliationschema, row counts, aggregates, row-by-row comparison
Key metric #3Zeb finished a NetSuite project in 2.5 weeks instead of 4

Picture a company whose entire analytics stack lives in a legacy warehouse such as Teradata or Oracle: hundreds of tables, loading pipelines, and chunks of SQL written years ago. To build AI applications and agents on that data, everything has to be rewritten for the new Databricks platform, and reports must show the same numbers after the move. This usually takes a whole engineering team several quarters, so the migration gets postponed indefinitely.

Cognition, the maker of the AI engineer Devin, has launched Devin for Databricks Migrations together with Databricks. According to the companies, a Fortune 500 energy company moved a platform of 142 objects (units of data and code) into Unity Catalog, Databricks' unified catalog with centralized access control, in just eight days. If you have a "move" like this that has sat in the queue for years, this news is about exactly that.

What Devin does during a migration

The process is familiar to anyone who has migrated from Teradata, Oracle, SQL Server, Informatica, Redshift, or Snowflake. You rewrite the legacy warehouse code for Databricks SQL, PySpark, or Spark Declarative Pipelines, run it, work through what fails, and verify that the data matches. Devin takes over this entire cycle: it converts the code, tests it directly in the customer's workspace, fixes what fails, and before cutover reconciles every source against its copy in Databricks.

Reconciliation works in four layers: table schema, row counts, aggregates (sums and similar totals), and row-by-row comparison. This is the most important part. An accountant who finds a discrepancy in a report a month after the move is the classic migration nightmare, and automated pre-launch checks target exactly that.

The main source of speed is parallelism. According to Cognition, Devin can autonomously run hundreds of tasks at once: a migration involves dozens or hundreds of independent objects, and each can be handed to a separate agent instance. That is where the eight days for 142 objects comes from.

Zeb's numbers and the economics

The second example is more modest, and therefore more realistic. The consulting company Zeb migrated data from NetSuite to Databricks for an international investment consulting firm: Devin and Zeb engineers built the pipelines and 20 tables. The project, planned for four weeks, took two and a half, roughly 37% faster. Zeb CEO Mal Vivek says Databricks delivery teams are now paired with Devin because the customer gets "more value per dollar."

The announcement says nothing about pricing or token consumption. All that is known is how to buy it: the solution is available in Databricks Marketplace, and payment can be applied against an existing Databricks Universal Commit contract. For a finance department, this simplifies things: no separate contract is needed. There are three ways to work with it: with your own engineers, with Databricks field engineers, or through partners such as Zeb. Devin is also named as one of Databricks' preferred migration tools.

There is also an infrastructure nuance. Cognition calls itself a strategic inference provider for Databricks: a range of open and frontier models is available through a single endpoint, Unity Gateway. This lets teams pick a model for each task and swap it without rebuilding the stack. Databricks head Ali Ghodsi adds that through Devin and Unity Gateway, context from legacy systems moves into Databricks under unified access control.

What to keep in mind

The announcement is a vendor publication. There are no independent measurements, the Fortune 500 customer is not named, and 142 objects is an impressive but not gigantic platform. In warehouses where stored procedures with undocumented business logic have piled up over the years, complexity grows nonlinearly, and data reconciliation is no substitute for understanding why that logic is needed. Zeb's 37% gain is a more honest benchmark than eight days.

The practical takeaway: if a Databricks migration is already approved but sits in the queue behind product work, it makes sense to pilot it on one isolated data domain. Measure three things: the share of objects that pass reconciliation without human edits, review time, and total cost. Human review, not code conversion, will most likely determine the real savings.

Story

Key Questions Answered

What Devin does during a migration?

The process is familiar to anyone who has migrated from Teradata, Oracle, SQL Server, Informatica, Redshift, or Snowflake. You rewrite the legacy warehouse code for Databricks SQL, PySpark, or Spark Declarative Pipelines, run it, work through what fails, and verify that the data matches. Devin takes over this entire cycle: it converts the code, tests it directly in the customer's workspace, fixes what fails, and before cutover reconciles every source against its copy in Databricks.

What are the key details on Zeb's numbers and the economics?

The second example is more modest, and therefore more realistic. The consulting company Zeb migrated data from NetSuite to Databricks for an international investment consulting firm: Devin and Zeb engineers built the pipelines and 20 tables. The project, planned for four weeks, took two and a half, roughly 37% faster. Zeb CEO Mal Vivek says Databricks delivery teams are now paired with Devin because the customer gets "more value per dollar." The announcement says nothing about pricing or token consumption.

What to keep in mind?

The announcement is a vendor publication. There are no independent measurements, the Fortune 500 customer is not named, and 142 objects is an impressive but not gigantic platform. In warehouses where stored procedures with undocumented business logic have piled up over the years, complexity grows nonlinearly, and data reconciliation is no substitute for understanding why that logic is needed. Zeb's 37% gain is a more honest benchmark than eight days.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…