DevOps Meetup #87: AI & Kubernetes

Talk
Stop Burning Tokens: Feedback Loops for AI Agents in Kubernetes

Summary
Here's something we learned the hard way: AI coding agents are great at generating code, but terrible at knowing whether that code actually works. Without a way to verify changes against real dependencies, agents just keep guessing, burning tokens on reasoning steps that never converge.
Everyone says the answer is better prompts. Send less context, constrain the output, and start fresh conversations. That helps, but it misses the bigger picture. The real waste comes from agents chasing integration bugs they have no way to see.
In this talk, I'll share how we ran into this at MetalBear while using agents for frontend work and how we fixed it by doing something simple: letting the agent run our existing E2E tests against real Kubernetes services using mirrord. No new tests, no mocks, no deploying after every change. Just a fast feedback loop that tells the agent "that worked" or "that broke" in seconds.
I'll walk through a real example, show what changes when agents can self-correct, and discuss how this works when your whole team is doing it simultaneously.

Speaker
Jake Page
Former high-school teacher turned DevOps engineer and now DevRel engineer at MetalBear, the company behind mirror. Over the past 6 years, he has focused on cloud-native developer tooling across packaging, software delivery, FinOps, testing, and debugging, with a strong Kubernetes angle throughout. Now based in Lisbon, Jake helps engineering teams adopt better development and testing workflows for Kubernetes services, writes about the intersection of AI tooling, developer experience, and cloud-native infrastructure, and frequently speaks at meetups and conferences.

Agenda
18:30 - 19:00 Opening
19:00 - 19:30 Welcome Coffee & Snacks
19:30 - 20:30 Stop Burning Tokens: Feedback Loops for AI Agents in Kubernetes by Jake Page

Location
Aubay Portugal at Porto
Av. da Boavista 1180 3º piso, 4100-113 Porto

Where to find us
Web: devopsporto.com
LinkedIn: linkedin.com/company/devopsporto
X: x.com/DevOpsPorto
Slack: devopsporto.slack.com

Calls
Organizers: tinyurl.com/callfororg
Proposals: tinyurl.com/callforprop