AI NewsProductivityAnnouncement
Vodafone Czech Republic says Gemini agent cut data pipeline costs 30 percent
Google Cloud says Vodafone Czech Republic used its Data Engineering Agent to rewrite legacy BigQuery procedures. It says the new pipelines cost 30 percent less and run twice as fast.

Why it mattersA measured 30 percent cost drop on a real production pipeline is more useful than a vendor demo. Readers still have to verify it on their own workload.
A data team can see a Gemini agent demo and still not know what it does on a real pipeline. Google Cloud published numbers this week from Vodafone Czech Republic. The carrier used Google's Data Engineering Agent to rewrite old BigQuery stored procedures. Google Cloud says the new pipelines cost 30 percent less to run.
The Data Engineering Agent is a Gemini tool inside BigQuery and Dataform. Vodafone used it to turn legacy stored procedures, some 2,000 or 3,000 lines long, into modular Dataform pipelines. The post is co-written by Helena Guerra, a senior manager at Vodafone, and Pedro Esteves at Google Cloud. No specific pipeline is named.
The numbers Google Cloud gives
Google Cloud says the new pipelines are 30 percent cheaper to run. It says they run twice as fast, a 50 percent cut in execution time. It says the migration itself was 90 percent faster than doing it by hand. Google Cloud does not say how many procedures were migrated, or against what baseline the execution time is measured.
The team asked the agent to parallelise the pipeline. It then asked the agent to cache intermediate tables in BigQuery materialized views. That cache is where the 30 percent saving comes from. Running steps in parallel can defeat BigQuery's own query cache. Materialized views put the cache back.
Where it is going next
Vodafone plans to roll the same approach out to its operations in Romania, Ireland, Germany and the United Kingdom. Google Cloud does not say when. The post also lists other patterns the team used. The agent handled dependencies between steps on its own. It wrote execution logs that replaced a spreadsheet the team had maintained by hand. It can run only part of a pipeline for debugging.
What a data team can take from this
Three figures on one real pipeline, all from the vendor and the customer together. A data team can read them as a starting point, not as a benchmark they can run. The agent was tested on serial workloads. These were long stored procedures with repeated subqueries. A workload that is already parallel and materialised will have less to gain.
The cost saving is the sharpest claim, because it ties to a named cause: cached intermediate tables in materialized views. That is checkable with a before and after bill on the reader's own project.
Everything here is Google Cloud's own writeup with Vodafone as co-author. The figures are the companies' own. No outside benchmark is given. A team moving a serial stored procedure into Dataform can use this as a target. The check is their own query bill in the same billing period, before and after.
Source
How Vodafone is using the Data Engineering Agent to improve their data pipelines, Google Cloud blog.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.

