From Tokens to Features: Architecting Cost Attribution for AI-Assisted Engineering | Harness Resource | Harness

AI coding tools are now widely deployed, but engineering organizations often lack the telemetry needed to determine whether that investment is improving delivery speed, quality, or cost. This session explores the architecture behind AI-assisted engineering measurement: connecting LLM usage and spend to developer sessions, commits, pull requests, tickets, and ultimately delivered features. It examines the data and instrumentation required, why provider-level usage data often lacks the necessary granularity, and how to build a vendor-neutral measurement layer for evaluating AI coding tools and engineering outcomes. Attendees will leave with a practical framework for designing an architecture that connects AI consumption to engineering activity—and enables meaningful comparisons of cost, productivity, and outcomes across tools and workflows. About the speakers: Martin Reynolds is a technology leader and speaker with over 30 years in software development, architecture, and delivery. Specializing in DevOps, DevSecOps, and DevEx, he has been featured in The New Stack, BeatNews, and LeadDev, and helps organizations balance speed, quality, and security in software delivery. Barani Mohandoss is Director, Cloud Governance, FinOps & E-Tools at Trimble. He has led enterprise adoption of AI coding tools like Copilot, Cursor, and Claude, building the cost visibility that ties AI spend to business outcomes. He focuses on making sense of token spend and value when you are both consumer and provider.