The Telemetry Trap — and How to Avoid It

Brian Mullins, Head of Engineering at Diaceutics, has a simple philosophy: cost data should live where engineering decisions get made. In this episode, he walks through how his team used granular telemetry and unit cost testing to double platform usage while holding costs flat — and how that same discipline now governs every AI workload they run. From a $6K/month redundant job to catching a misfired warehouse tier, the examples are specific and the playbook is replicable.

This episode covers how Brian’s team doubled platform query volume while holding costs flat by applying the same cost visibility and attribution practices to AI workloads that they already had in place for traditional cloud infrastructure. He walks through real examples: eliminating $6,000/month in redundant processing, catching a warehouse left on the wrong tier, and using unit cost experiments at 10, 100, and 1,000 records to project AI spend before it scales. He also makes the case that cost anomalies are not just a FinOps signal — they are a security signal too.

Key Takeaways

AI spend is just another line item — treat it with the same tagging, attribution, and anomaly detection discipline you already apply to everything else.
Unit cost experimentation at small scale is the most reliable way to project AI costs before they run away from you.
High cost is not always wrong cost — sometimes architecture decisions justify the spend. The goal is knowing which is which.
Cost visibility adoption spreads faster when it’s embedded in existing team processes, not introduced as a separate tool or mandate.
A healthy paranoia about cost spikes doubles as a security posture — unexplained cost is always worth a second look.

About the guest

Brian Mullins leads engineering at Diaceutics, where his team builds a platform that helps connect patients to the right diagnostics and treatments. He has embedded FinOps discipline deeply into engineering workflows – covering multi-cloud cost attribution, AI workload tracking, and telemetry-driven optimization. His approach combines infrastructure precision with a cost-benefit mindset that extends well beyond the engineering team.