Abstract
99.9% uptime. Zero pod restarts. p99 latency looking fine. And still, a quarter of your users never finished onboarding that week.
Good uptime tells you your servers are running. It tells you nothing about whether your product actually works for the people using it. And the faster you ship, the more that gap costs you. AI-assisted development makes this even more urgent. We write and deploy code faster than ever now. But AI doesn't know your users. It doesn't know what makes your signup flow fragile or why your invoice job silently drops requests at the end of a quarter. That context is yours to build and protect.
Observability-Driven Development is a practice I've been refining in Go production systems. The core idea is simple: you define how to measure a feature before you write it. The SLI comes first. Not as an afterthought, not after something breaks in production. Before the code.
In this talk I'll show how that works in practice. How to instrument Go services so that when something goes wrong, you move from symptom to cause in minutes instead of days. How to write SLOs that Product and Engineering actually share ownership of rather than argue about. And how to build release guardrails that catch regressions before your users feel them.
We'll also get into what this means for agentic workflows. When AI agents run parts of your system autonomously, the usual "we'll add monitoring later" approach breaks completely. Agents fail quietly. They complete tasks in unexpected ways. Without instrumented user journeys as the ground truth, you have no way to know if an agent did the right thing or just the technically correct thing. ODD gives you that anchor.
What you'll take away:
- A concrete workflow for defining SLIs before writing code
- Practical Go instrumentation with OpenTelemetry that ties traces to user journeys not just service health
- How to move from a metric change to a trace to a log to a real user impact in one flow
- SLO and error budget patterns built for shared Product and Engineering ownership
- Why agentic engineering raises the stakes and how ODD keeps you in control
Who it's for: Go developers who want production confidence from day one. No observability background needed.
