Lovable’s AI Agent Automates Much of Its Data Analysis

As Lovable grew, its in-house capacity for data analysis could not keep up. The company built Lovsight, an AI agent that investigates questions about business data, flags changes and suggests actions. Lovable’s head of Data Insights, Jonas Björk, described the system at Google Cloud AI Day in Stockholm.

Lovable hired its first data scientist when it had about 150 employees, but the demand for analysis soon exceeded what one person could handle. Lovsight combines language models with an internal AI gateway, instructions, business context, tools and memory. It can also draw context from internal systems and consult other agents with specialized knowledge. The company adds tools as new needs arise and uses “skills” to guide the agent through recurring tasks.

The agent follows a set process: it receives a problem, assembles a prompt, reasons with a language model, selects tools and reviews the results. If it has not met its goal, it repeats the process. Lovable sets limits on data access and tool use after seeing runs make hundreds of calls without useful results.

According to Björk, Lovable sends about 500 million events a day to BigQuery, and its data environment handles about 100,000 queries daily. Agents monitor internal KPIs and investigate changes by checking deployments, code changes, documentation and production logs. They report possible causes, supporting evidence, unresolved questions and suggested actions.

For September, Lovsight reported 167,000 completed jobs and more than 200,000 language-model calls; 99.56% of jobs finished without errors, according to the figures Björk shared. Lovable also uses agents to find customer projects that may be worth highlighting.