Reliability Rebels, Ep 10: Kyle Forster
Traditional observability data was built for humans looking at dashboards. ๐ช๐ต๐ฎ๐ ๐ฎ๐ฏ๐ผ๐๐ ๐๐ต๐ฒ๐ป ๐๐ ๐ฎ๐ด๐ฒ๐ป๐๐ ๐บ๐ฎ๐ธ๐ฒ ๐ฑ๐ฒ๐ฐ๐ถ๐๐ถ๐ผ๐ป๐?
In my latest Reliability Rebels episode, Kyle Forster (former Sr. Director of Kubernetes at Google, now CEO of RunWhen) shares data that challenges our assumptions about what agents actually need:
๐ช๐ต๐ฒ๐ป ๐ฅ๐๐ป๐ช๐ต๐ฒ๐ป'๐ ๐ฐ๐๐๐๐ผ๐บ๐ฒ๐ฟ๐ ๐ฝ๐ฒ๐ฟ๐ณ๐ผ๐ฟ๐บ ๐๐-๐ฑ๐ฟ๐ถ๐๐ฒ๐ป ๐ถ๐ป๐๐ฒ๐๐๐ถ๐ด๐ฎ๐๐ถ๐ผ๐ป ๐ผ๐ณ ๐๐ต๐ฒ๐ถ๐ฟ ๐ฝ๐ฟ๐ผ๐ฑ๐๐ฐ๐๐ถ๐ผ๐ป ๐๐๐๐๐ฒ๐บ๐, ๐บ๐ฒ๐๐ฟ๐ถ๐ฐ๐, ๐น๐ผ๐ด๐, ๐ฎ๐ป๐ฑ ๐๐ฟ๐ฎ๐ฐ๐ฒ๐ ๐ฎ๐ฐ๐ฐ๐ผ๐๐ป๐ ๐ณ๐ผ๐ฟ ๐ผ๐ป๐น๐ ๐ฏ๐ฌ% ๐ผ๐ณ ๐๐ต๐ฒ ๐๐ผ๐ผ๐น ๐ฐ๐ฎ๐น๐น๐ ๐บ๐ฎ๐ฑ๐ฒ.
The other 70%? Context. Configuration data. Safe DB calls. API health checks. The digital equivalent of a senior engineer "wandering around the system and poking around" before an incident even happens.
Kyle's bet: By end of year, we'll stop building observability for human eyeballs and start building it for agentic loops- thousands of narrowly scoped skills that retrieve specific context, not mountains of retention you query after the fact.
Orders of magnitude less data. Richer information. ๐ก๐ผ ๐ด๐ถ๐ฎ๐ป๐ ๐ฟ๐ฒ๐๐ฒ๐ป๐๐ถ๐ผ๐ป ๐ฏ๐ถ๐น๐น.
We also dig into his personal encounter with the "AI code tsunami," what happens when velocity massively outpaces review capacity, and what "operational readiness" means when teams operate systems they didn't write.
I love my infrastructure dashboards. It's uncomfortable to consider they might become less relevant- but I can't ignore Kyle's predictions.
Links: