How CosmoLaser's Chief Marketer Uses Claude Code to Run His ETL Pipelines on Tower

In the last six months, AI coding assistants like Claude Code have shifted from experimental tools to practical infrastructure. Since Claude Code became widely available in mid-2025, along with Anthropic’s Opus 4 and Sonnet 4 models, many teams now rely on AI assistants to build and maintain real data systems.

What sets Claude Code apart is its ability to support every stage of software development. It can interpret business needs, draft implementations, run tests, inspect production logs, and fix errors. For small teams without a dedicated data platform function, this is especially powerful. But even with AI handling much of the coding, teams still need a reliable way to run that code in production. That’s where Tower comes in.

Morten Nielsen, Founder and Chief Marketing Officer at cosmolaser.dk, a chain of cosmetic treatment clinics in Denmark, has embraced this shift.

Nielsen built a data platform that consolidates seven sources, including Google Analytics, into a BigQuery warehouse. The platform gives him visibility into clinic performance and supports both daily operations and long-term planning.

He chose dltHub for data ingestion and dbt for transformations. With initial guidance from advisors at Pyne, he now runs and improves the system largely on his own, using Claude Code to build and evolve the pipelines. When it came time to put his dbt models and dltHub pipelines into production, he chose Tower to make them reliable and repeatable.

Nielsen explains:

We rely on Claude to generate and evolve our data pipelines, integrating multiple operational data sources into our BigQuery data warehouse. Tower provides the runtime and structure needed to turn that AI-generated code into reliable, production-ready pipelines. It removes the operational overhead and lets us run modern dbt and dlt workflows with a small, fast-moving team.

In this Tower demo, Claude is asked to build a data agent that answers questions about a dataset. It creates an initial version, deploys it, encounters a dependency error, checks the logs, fixes the issue, and redeploys. The cycle repeats quickly until the data agent runs successfully. 

Coding agents building data agents. It’s very meta.

Nielsen’s story reflects a broader shift. People who didn’t previously consider themselves engineers are now building and operating real data systems. As AI handles more of the coding and iteration, the core question has changed from “Can I write this?” to “Can I run this reliably in production?”

Tower is seeing the same pattern across other industries, including software and finance, where teams are building AI-powered data tools. In the past 30 days alone, the Tower SDK has been downloaded more than 50,000 times - evidence of its growing role in modern data engineering.

To support teams running dbt Core on Tower, we recently released a dbt extension in the Tower SDK. The Tower dbt solution page includes additional resources, including a complete dbt example app.

For business operators like Nielsen who want to add Tower to their AI toolkit, there’s also a five-minute Quickstart that shows how to use Claude with Tower MCP. It’s a straightforward way to see what your own AI-built pipelines look like running reliably in production on Tower.