Uploaded April 2026 | Updated September 2026, 3 days ago
In this episode from the WorkOS booth at HumanX 2026, Michael Grinich sits down with Christine Yen, co-founder and CEO of Honeycomb, one of the defining observability platforms in engineering, built before the AI wave and now at the center of it.
Christine's key observation: as AI coding agents ship code faster than humans can read it, the codebase stops being the source of truth for understanding what a system does. The telemetry becomes the record. That changes everything about how software teams debug, understand, and operate their systems.
Key topics covered:
• Why the codebase is no longer the source of truth when no one has time to read it
• Why you need to capture not just what happened, but the inputs that led to a decision
• The difference between evals and observability: evals are for development, observability is for production, and they should feed each other
• Why there's no such thing as "AI observability" vs. "traditional observability" — it's all just software
• How a 20-year-old company doubled its custom code in two years, and what that means for visibility
• SLOs as a way to encode what "good" means — and how that becomes a reward function for agents
• What it looks like when observability feeds directly back into the development loop
Learn more about WorkOS: workos.com and Honeycomb: honeycomb.io
In this episode from the WorkOS booth at HumanX 2026, Michael Grinich sits down with Christine Yen, co-founder and CEO of Honeycomb, one of the defining observability platforms in engineering, built before the AI wave and now at the center of it.
Christine's key observation: as AI coding agents ship code faster than humans can read it, the codebase stops being the source of truth for understanding what a system does. The telemetry becomes the record. That changes everything about how software teams debug, understand, and operate their systems.
Key topics covered:
• Why the codebase is no longer the source of truth when no one has time to read it
• Why you need to capture not just what happened, but the inputs that led to a decision
• The difference between evals and observability: evals are for development, observability is for production, and they should feed each other
• Why there's no such thing as "AI observability" vs. "traditional observability" — it's all just software
• How a 20-year-old company doubled its custom code in two years, and what that means for visibility
• SLOs as a way to encode what "good" means — and how that becomes a reward function for agents
• What it looks like when observability feeds directly back into the development loop
Learn more about WorkOS: workos.com and Honeycomb: honeycomb.io
