TL;DR: AI is accelerating code development, which makes waiting for realistic test data the new bottleneck. Database branching shortens that wait to seconds by letting teams create isolated, production-like environments without lengthy copying and provisioning. But as environments get easier to create, governance matters more than ever, as branches can still contain production-derived data. Pairing Databricks Lakebase with Quest Lakebase Modeler helps teams combine the speed of branching with the modeling, validation, and control enterprises need.

AI has shifted the development bottleneck from writing code to validating it

The world’s largest AI companies are all pointing toward the same trend: software development is becoming less about writing code by hand and more about reviewing, validating, and guiding AI-generated output.

Google CEO Sundar Pichai reported that nearly 75 percent of new code at Google is AI-generated and approved by engineers. Microsoft CEO Satya Nadella has estimated that AI was writing roughly 20 to 30 percent of the code in some Microsoft repositories and projects. And Anthropic CEO Dario Amodei has predicted that AI will soon write the vast majority of code.

Whether your teams use GitHub Copilot, Databricks Assistant, ChatGPT, Claude, or purpose-built coding agents, AI is changing the pace of development. But faster code generation still requires validation. Every AI-generated change needs testing against realistic data before it reaches production, and that’s where one of software delivery’s most stubborn bottlenecks sits. As development cycles accelerate, test data management and environment provisioning can struggle to keep pace.

The hidden cost of waiting for test data and environments

Ask developers what slows them down and you’ll hear familiar answers: waiting for environments, refreshing stale data, fighting over shared staging systems, reproducing production issues, and coordinating database changes across teams. These bottlenecks evolved as applications grew more complex and governance requirements tightened.

A traditional provisioning process often looks like this:

  • A developer requests a test environment
  • Infrastructure or database administrators approve it
  • Someone creates a production copy
  • Sensitive data gets masked
  • Access is provisioned
  • Testing finally begins

Each step makes sense on its own. Together, they add up to friction.

Waiting a day for a test database might once have been an inconvenience. In an AI-assisted workflow, it can become the rate-limiting step for the entire delivery process. Developers can generate code in minutes, but that code sits idle until teams can test it.

The costs add up, too. Database copies drive up infrastructure spending, shared environments create contention between teams, and every production-derived environment raises compliance and security questions.

Teams end up juggling competing goals: realistic data can expose sensitive information, stronger privacy protections can reduce test fidelity, and infrastructure costs can limit how many environments are available.

Now consider what happens when dozens of developers, CI/CD pipelines, automated test frameworks, and AI agents all need isolated access to realistic data. More code means more schema changes, more pull requests, more integration tests, and more temporary environments. Traditional shared staging environments can become a significant constraint.

This growing need for fast and reliable validation is one reason database branching is drawing so much interest from development, platform engineering, and data teams.

What is database branching and how does it work?

Database branching works much like branching in source control but applies the concept to databases. A branch is an isolated, writable version of a database that teams can create quickly, modify independently, and use to validate changes before promoting approved work or discarding the test environment.

Traditional environments often scale through duplication. Need another environment? Copy the database. Need to validate a feature branch? Copy it again. The model works until the economics become difficult to sustain: storage grows, provisioning effort increases, and refresh cycles get more complicated.

Databricks Lakebase takes a different approach. It uses copy-on-write branching, so branches share underlying storage with their parent and store changes separately as data is modified. According to Databricks’ overview of database branching, this lets teams create isolated, writable branches without duplicating the parent database.

From a developer perspective, the workflow is remarkably simple:

  • Create a branch
  • Make changes
  • Run tests
  • Apply approved changes through the deployment pipeline or discard the branch

What previously required substantial database cloning operations can now happen in seconds, making temporary validation environments much more practical.

That unlocks something teams have wanted for years: an isolated environment for every developer, pull request, test cycle, or AI agent. Developers spend less time competing for shared resources. Validation can happen earlier and against more realistic data. Experiments become easier to isolate, and unsuccessful changes can be discarded without affecting the source database.

Database environments shift from scarce resources to flexible, disposable ones, and that’s a significant change. But this doesn’t eliminate the need for governance.

Database branching makes data governance more important

The biggest misconception about database branching is that environment creation is the hard part. In enterprise environments, governance is often the harder problem.

Branching dramatically improves availability, but a branch created from production can still contain production-derived data.

So, you still need answers to the same questions:

  • Is sensitive information protected?
  • Which masking policies apply?
  • Who gets access?
  • How do you maintain audit requirements?
  • How do changes move through environments?

When environments get easier to create, governance matters more. Faster provisioning without appropriate controls can multiply risk.

The strongest approach starts with governed source environments that already have the right protections. Organizations can apply masking, generate synthetic data, remove regulated information, or restrict sensitive attributes before creating downstream branches.

Branching from a governed source helps teams carry those protections forward while maintaining the agility of rapid environment creation. But verifying that the right controls remain in place is still essential.

How Quest Lakebase Modeler adds governance to Lakebase branches

Databricks Lakebase provides the speed and flexibility to create branches on demand. Quest Lakebase Modeler complements that capability by helping teams design, review, validate, and prepare database changes for controlled deployment. It runs inside the customer’s Databricks workspace and is available through the Databricks Marketplace.

The workflow follows four key stages:

Create. Start from business requirements, existing databases, Unity Catalog assets, or a new design. This establishes a modeling foundation that helps keep business intent and technical implementation aligned.

Enrich. AI-assisted recommendations suggest relationships, domains, descriptions, and metadata. These remain recommendations subject to human review, so teams retain control over approved changes.

Map. Connect logical designs to Unity Catalog assets or translate them into deployable physical structures. This helps bridge the gap between business requirements and database implementation.

Deploy. Translate approved models into deployment-ready changes that teams can validate in a Lakebase branch before promoting them through the production deployment process.

In practice, teams can design a change, generate the deployment, apply it to an isolated branch, and validate it against production-like data before deciding whether to promote the approved changes.

If validation fails, teams can discard the test branch without affecting production. That reduces the need for complex rollbacks and environment rebuilds, making it easier to start fresh.

Together, Lakebase and Quest Lakebase Modeler connect faster database environment provisioning with a more structured approach to modeling, change validation, and deployment governance.

Why this matters for AI-era software delivery

For years, organizations built development processes around scarce infrastructure. Database copies were expensive, provisioning took time, and environments were limited, so teams adapted. AI is changing those assumptions. As development throughput rises, testing and validation infrastructure need to scale with it.

Database branching addresses one of the biggest sources of friction in software delivery: access to realistic, isolated environments.

And the benefits extend across the organization:

Developers spend less time waiting for test environments

Platform teams manage fewer full database copies and reduce provisioning overhead

Governance teams can establish more consistent processes for reviewing and validating changes

Business leaders can accelerate delivery while maintaining appropriate compliance controls

AI is changing how code gets written. Database branching is changing how that code gets validated. Teams that can provision production-like environments in seconds while maintaining effective governance controls are better positioned to accelerate delivery while keeping risk in check.

Mark Gowdy is Chief Partner Technologist at Quest, leading strategy and co‑innovation with top technology partners across Data, Cybersecurity, and Platform Modernization. With 25+ years in data management, observability, and cloud optimization, he helps global enterprises build trusted data and security foundations for AI success.

See Quest Lakebase Modeler in action

Watch experts from Databricks and Quest demonstrate how Quest Lakebase Modeler and Databricks Lakebase work together to accelerate testing while maintaining governance and control.