Vodafone Czech Republic has used Google’s Data Engineering Agent to modernise data pipelines built around legacy BigQuery stored procedures, with Google Cloud reporting that pipeline execution times were cut by half. The customer example, published on November 20, shows how AI agents are being positioned for practical data-engineering work rather than only code generation or chat-based assistance. The project is described by Google Cloud, so performance and cost claims should be treated as vendor-reported results for this deployment, not as a guarantee that every BigQuery workload will see the same improvement.
The work involved converting stored procedures into more modern data pipelines. Stored procedures can be useful for centralising business logic, but older implementations may become difficult to maintain as data volumes, dependencies and reporting requirements grow. A modern pipeline can make changeations easier to test, schedule and observe, depending on the architecture. Google’s customer story describes the use of its Data Engineering Agent in this modernisation effort and reports a reduction in execution time. The original reporting on Vodafone sets out the primary development.
The project focused on legacy data changeation work
The reported performance improvement is meaningful because data pipelines often sit between operational systems and analytics platforms. Slow changeations can delay dashboards, customer insights and downstream applications. They can also increase compute costs when jobs run longer than necessary or require repeated processing. However, execution time is only one measure of a pipeline’s quality. Teams also need to consider correctness, reliability, data freshness, maintainability and the cost of operating the new workflow. A faster changeation that produces inconsistent results would not be a successful modernisation.
AI-assisted engineering can help by analysing existing procedures, identifying changeation logic and proposing a new structure. But legacy data code often contains business rules that are not documented elsewhere. An agent may be able to translate syntax while missing assumptions about time zones, null values, rounding, historical corrections or how departments interpret a field. Developers need to compare outputs from the old and new pipelines using representative datasets and edge cases before replacing a production process. Further technical context for Google Cloud is available in the supporting documentation.
Benchmark the old and new pipeline against the same data
Vodafone’s example is relevant to large enterprises because data estates frequently contain years of accumulated code. Organisations may have built separate pipelines for billing, customer operations, network performance and finance. Rewriting everything manually can be expensive, while leaving the code unchanged can make it harder to respond to new analytics and AI requirements. An agent can reduce the effort of inspecting and changeing code, but the project still requires data owners and engineers who understand the underlying business process. The AWS data-quality capability is a related part of the modern data stack, covered in AWS Adds Catalog-Level Data Quality Monitoring to Glue Data Catalog.
A data-engineering agent should be evaluated against a reproducible baseline. Teams can capture the current pipeline’s run time, compute consumption, failure rate and output checks, then compare those measures after migration. They should also retain the original procedure until the replacement has completed a representative period of production use. This is important when pipelines contain business rules that are not documented in a central specification. A faster workflow is useful only when it preserves the meaning of the data and remains supportable after the initial modernisation project ends.
The reported gains depend on the workload
The use of an agent also introduces governance questions. A data-engineering system may need access to schemas, stored procedures, sample records and deployment environments. Organisations should limit the data it can read, separate development from production and ensure that generated changes are reviewed before deployment. If customer information is involved, teams should use approved test data and apply the same privacy controls they would use for any external development tool. AI assistance should not create a shortcut around data-access policies. The source coverage of BigQuery provides the underlying detail.
Google’s report says the agent is available within its data-engineering workflow, but the precise results will depend on the structure of each customer’s environment. A pipeline with clear schemas and well-tested changeations may be easier to modernise than one with undocumented dependencies and inconsistent source data. Organisations should begin with a contained workload, establish baseline execution time and cost, compare output quality and document the review process. That provides evidence for deciding whether the agent can be used on a larger set of pipelines.
The Vodafone case shows a shift in enterprise AI from generating isolated code snippets to assisting with multi-step engineering tasks. The reported time reduction is encouraging, but the durable value will be measured by how reliably the new pipelines operate after the project team moves on. Companies should evaluate performance alongside correctness, observability and maintainability. AI agents can accelerate data modernisation, but they work best when engineers retain control over requirements, validation and production changes.