How CGI engineers helped a healthcare organization accelerate AI-assisted software development with Databricks while maintaining enterprise governance and open-source portability.

Enterprises want AI acceleration without sacrificing future choice

Modern AI platforms are helping organizations accelerate software development, strengthen governance and improve engineering productivity. At the same time, many organizations are asking a practical question: how can they take advantage of these platforms without creating unnecessary technology dependencies or increasing the risk of vendor lock-in?

The challenge for organizations operating in highly regulated industries extends beyond productivity; they also need to protect sensitive data, maintain rigorous governance and preserve the flexibility to adapt as technology evolves.

For one healthcare utilization and care management client, maximizing the value of its Databricks platform also meant enabling long-term architectural flexibility. CGI engineers designed an AI agent sandbox that supported autonomous software engineering while preserving enterprise governance and continuously validating compatibility with open-source technologies, helping reduce dependence on a single platform over time. The result was an engineering environment where AI agents could independently develop, test and validate software changes before handing them to developers for final review and deployment.

The challenge

The client's data platform was already built on Databricks and Unity Catalog, providing enterprise governance, centralized metadata management, scalable Spark execution and accelerated data engineering. As the organization explored autonomous software engineering to reduce development cycle times, it also wanted confidence that future technology decisions would not be constrained by proprietary dependencies.

The challenge: how could they leverage the full capabilities of Databricks while maintaining a clear path to open-source portability?

The solution: AI agent sandbox

Working with the client, CGI engineers designed an ephemeral AI agent sandbox built on open-source technologies including Apache Spark, open-source Unity Catalog and standard development tools. AI agents could develop, test and validate software changes in isolated environments while continuously validating compatibility with open-source technologies.

Equally important: no production healthcare data entered the sandbox. Only metadata structures and Unity Catalog definitions were synchronized, allowing AI agents to work in a representative development environment without exposing regulated PHI or PII.

Governance before autonomy

One of the defining design principles of the solution was separating deterministic processes from autonomous decision-making.

Everything that prepared the development environment remained deterministic, including sandbox creation, metadata synchronization, repository setup and environment configuration. Only after those governed processes were complete did autonomous agents begin making engineering decisions.

This approach supported repeatability, auditability, governance and compliance while reducing operational risk without limiting the benefits of AI-assisted engineering.

AI agents with independent responsibilities

Once the sandbox was prepared, the development agent took ownership of the user story, by:

  • Interpreting functional requirements
  • Implementing code changes
  • Generating unit tests using synthetic data
  • Executing automated validation before committing successful changes to a feature branch

Rather than validating its own work, the development agent then handed responsibility to a separate QA agent operating in its own isolated sandbox. The QA agent generated independent validation scenarios instead of reusing developer-created tests, verifying functional requirements, regression impacts, error handling and boundary conditions from a fresh perspective.

The development and QA agents continued iterating until both independently agreed the implementation satisfied the defined acceptance criteria. This created an autonomous quality feedback loop before software reached a human reviewer, strengthening software quality while reducing manual effort.

Human oversight remains essential

Despite the high degree of automation, final accountability remained with engineers. Following successful agent validation, code was automatically prepared within the Databricks development environment for review using existing enterprise governance processes before deployment.

This balanced AI autonomy with regulatory requirements and engineering accountability, an especially important consideration in healthcare environments.

Reducing vendor lock-in through continuous validation

The sandbox was intentionally built on open-source components, including Apache Spark, open-source Unity Catalog and containerized infrastructure. Every AI-generated code change was continuously validated against those technologies throughout development.

This helped keep applications aligned with portable architectural patterns while allowing the client to continue benefiting from Databricks' enterprise governance, unified data management, scalable execution and development productivity. By building portability into the engineering process, the organization maintained the flexibility to support future technology decisions while continuing to innovate on its existing platform.

Business outcomes

The AI agent sandbox delivered value across several areas:

  • Faster software delivery through AI-assisted implementation, automated testing and shorter engineering cycles
  • Higher software quality through independent AI validation, expanded test coverage and continuous feedback loops
  • Stronger data protection through metadata-only synchronization and isolated execution environments
  • Greater architectural flexibility by reducing vendor lock-in risk through continuous validation against open-source technologies
  • Higher-value engineering work by reducing time spent on repetitive coding and testing, allowing engineers to focus on architecture, design decisions and business priorities

While this solution was developed for a healthcare organization operating in a highly regulated environment, the engineering principles extend well beyond a single industry.

Organizations across sectors are looking for ways to accelerate AI-assisted engineering while maintaining governance, security and long-term technology flexibility.

Looking ahead

AI-assisted engineering is changing how enterprise software is developed, tested and delivered. The greatest opportunity lies not only in accelerating development, but in creating engineering environments where automation, governance and architectural flexibility work together.

By combining deterministic processes, independent validation, human oversight and open standards, organizations can adopt AI-assisted engineering with greater confidence while preserving the flexibility to evolve as technologies and business priorities change.