Flock Product & Engineering Blog
The Goldilocks Problem: Building Insurance Products That Are Just Right
Commercial fleet insurance has to fit both a London Uber driver and a national haulier. How we landed on typed product configuration sitting on top of a library of shared behaviours, and stopped treating speed and depth as a tradeoff.
Building Jay: Flock's Production Grade Agent
How an internal MCP experiment turned into Jay, an agentic AI assistant embedded in Flock's fleet portal that answers fleet managers' questions with live data, charts and maps.
Latest articles
The Goldilocks Problem: Building Insurance Products That Are Just Right
Commercial fleet insurance has to fit both a London Uber driver and a national haulier. How we landed on typed product configuration sitting on top of a library of shared behaviours, and stopped treating speed and depth as a tradeoff.
Building Jay: Flock's Production Grade Agent
How an internal MCP experiment turned into Jay, an agentic AI assistant embedded in Flock's fleet portal that answers fleet managers' questions with live data, charts and maps.
From Telemetry to aFNOL: How We Built Automated Crash Detection
How Flock's data and ML teams built a system that turns raw vehicle telemetry into verified crash notifications, combining XGBoost inference, an AI agent swarm on Bedrock, and event-driven AWS infrastructure.
Small teams, fast deploys: how we built a feedback loop that compounds
The software engineering practices that make speed and reliability compound at Flock: from trunk-based development and fast feedback loops to type-safe contracts and isolated services.
Inside Flock's AI-Powered Product & Engineering Workflow
How Flock's engagement team drove a 152% increase in weekly active usage by fundamentally changing how product and engineering work together with AI.
Your data should outlast your stack
How we migrated off Snowflake-as-everything, built a proper lakehouse on AWS, and made the whole thing queryable by an AI via MCP.
From hackathon prototype to production at Flock
How we turned a simple PDF extraction idea into a production-grade agentic pipeline that saves underwriters 3–4 hours every week.