How Class Editori Built MFGPT, Italy’s First Generative AI Media Agent

Class Editori has published financial and lifestyle news in Italy for four decades, operating newspapers, magazines, TV channels, and digital platforms under brands including Milano Finanza and Gambero Rosso. The company launched one of Italy’s first websites in 1996 and built itself into a major data broker for stock exchange information. But by the early 2020s, the competitive pressure had shifted: global platforms like Google, Amazon, and Netflix were training consumer expectations around personalized, conversational content discovery. A media house with 40 years of verified, high-quality journalism needed a way to deliver that archive in a format audiences now demanded.

The core challenge was fragmentation. Decades of content lived in disconnected systems — articles, podcasts, videos, and real-time market feeds — with no unified infrastructure to query across them. Building a conversational AI product on top of that required not just an AI layer but a complete rethinking of the data architecture. Class Editori also faced a compliance constraint: Italian law prohibited systems from crossing into personalized financial advice, so any AI agent needed guardrails built in from the start.

Working with Google Cloud partner Softlab, Class Editori built an end-to-end data and AI platform. BigQuery became the central warehouse for structured data, Dataflow managed complex ingestion pipelines, and Vertex AI provided a unified surface for both standard machine learning and generative AI model management. The flagship product to emerge from this platform is MFGPT, a multi-agent generative AI system built on Gemini Flash and Pro models. MFGPT lets subscribers query the full Class Editori archive — articles, podcasts, videos, and real-time financial data — through a conversational interface on the Milano Finanza website. It generates charts, runs company performance analyses, and uses Cloud Functions to set up automated personalized alerts, such as email notifications when new content on a specific topic is published.

The market response validated the approach immediately. MFGPT converted trial users into paid subscribers and opened B2B agreements with major financial boards. User feedback highlights that the agent outperforms general-purpose LLMs for financial analysis because of its depth of domain data and the quality of its infographics. The platform’s time-to-revenue was deliberately short: Softlab’s implementation methodology prioritized a production-ready launch over extended development cycles, allowing Class Editori to generate revenue from the technology before the project was even fully complete.

Class Editori is now treating MFGPT as the first node in a broader AI agent network. The team is expanding the platform to other publications, beginning with Gambero Rosso, where a lifestyle agent will incorporate map-based restaurant discovery. Internally, churn prediction models and dynamic paywalls are under development. The company’s stated ambition is to scale its initial success across the full breadth of its media portfolio, becoming a model for how legacy publishers can turn their content archives into AI-native products.