GLM5.2 and Kimi K3 have been great for my personal use cases Can't wait for Qwen 3.8.

"Building Service Topology at Scale: Architecture, Challenges, and Lessons Learned" by Parth Jain, Rakesh Sukumar, Yingwu Zhao, Renzo Sanchez-Silva and Nathan Fisher netflixtechblog.com/building-ser...

"The incident metrics mirage" by Brent Chapman greatcircle.com/blog/2026/05...

"AI in SRE: How Google is Engineering the Future of Reliable Operations" by Ioannis Papapanagiotou, Stevan Malesevic, Chris Heiser and Ruslan Meshenberg sre.google/resources/pr...

"Kimi K3, and what we can still learn from the pelican benchmark" by Simon Willison simonwillison.net/2026/Jul/16/...

Revision 182 is out! @koslib.com #devops #sre #platformengineering embracerisk.substack.com/p/revision-182

"Patterns for Deploying OpenTelemetry Collector at Scale" by Elizabeth Mathew signoz.io/blog/opentel...

"A Practical Guide to Reducing Token Spend" by Adam Jacob www.adamhjk.com/blog/a-pract...

Passes unit tests != Survives production. Real user traffic and behaviour, at scale, will surface issues very hard to anticipate. Building software is a learning process. As your customers grow and change and your software evolves, new and interesting challenges will emerge.

"OpenTelemetry Has Graduated… Now what?" by Adriana Villela, Reese Lee opentelemetry.io/blog/2026/ot...

"How k0rdent and Cilium Turn Kubernetes Sprawl into a Platform" by Christian Hernandez cilium.io/blog/2026/07...

As a leader, your words need to match your actions. If you say something is your top priority but you never have time for it and/or are lightly involved, people will infer it's not actually a priority. You need to make it clear why you're less involved so people believe it.

One goal of postmortems is to understand what went wrong so it doesn't happen again. But the most valuable part of it is that you learn and get to update the mental model you have of your systems. If you outsource this process, you learn nothing. You cannot outsource learning.

"Operating OpenTelemetry at scale with OpAMP" by Dotan Horovits www.cncf.io/blog/2026/07...

"When the Postgres query planner goes rogue" by Nick Van Wiggeren planetscale.com/blog/when-th...

"How Bullet uses ClickHouse Cloud to give DeFi's fastest exchange real-time analytics" by ClickHouse clickhouse.com/blog/bullet-...

Many tech organizations report the same: code reviews are the new bottleneck. Other industries have faced similar challenges: when output outpaces the capacity for review, the main method of accountability. How did other industries cope with that? mccricardo.com/bottlenecks-...

"Is a Pod the right deployment unit for an AI agent?" by Lin Sun www.cncf.io/blog/2026/07...

"Why killing pods is not enough: Testing operator reconciliation with operator-chaos" by Ugo Giordano developers.redhat.com/articles/202...

"Building Cross-Team SLO Contracts for Performance Accountability" by Ujjwal Gulecha dzone.com/articles/bui...

"AI Won't Keep You from Hitting the Scalability Wall" by Bru Woodring prismatic.io/blog/ai-wont...

"Argo CD 2026 User Survey Results" by Nikita 'Kit' Dergilev blog.argoproj.io/argo-cd-2026...

Revision 181 is out! @koslib.com #devops #sre #platformenginneering embracerisk.substack.com/p/revision-181

"You Can't Sample Your Way to Reliable Agents" by Vijay Karthik blog.oodle.ai/you-cant-sam...

A bad plan is better than no plan. The trick is to find how, as fast as possible, if you're on the right track or not and adjust accordingly.

"From Day 1 to Production: Building Lyft’s Analytics & Rides Intelligence Assistant as Onboarding Project" by Sagarbaronia eng.lyft.com/from-day-1-t...

"Network boundary for AI agents using NGINX and OpenTelemetry" by Marko Sluga www.cncf.io/blog/2026/07...

Just listened to two short podcasts during my lunch time. In one, the cost of producing code with LLMs was nearing zero. On the other, enterprises were cutting back because costs were too high. It's hard to have a clear picture of what is what.

"How we used AI agents to migrate GitLab rate limiting" by Sam Wiskow about.gitlab.com/blog/ai-agen...

"Purpose without power is just poetry. Give it teeth, and it becomes destiny" Eric Ries, Incorruptible: Why Good Companies Go Bad... and How Great Companies Stay Great

"The 4-body problem of SRE: Why autonomous operations depend on context" by Sanjeev Sharma www.cncf.io/blog/2026/07...

"Code maintainability plummets in the AI coding era" by Bill Doerrfeld leaddev.com/ai/code-main...

A beautiful characteristic of a complex system is that components can work wonderfully in isolation and fail spectacularly when combined. Failure is inevitable. Prepare for it.

"chDB as the Agent's Local Data Engine" by Auxten Wang and Nakul Mishra clickhouse.com/blog/chdb-ag...

"Helm upgrade --reuse-values Fails with Nil Pointer Error After a Chart Version Bump" by Dean Lewis veducate.co.uk/helm-upgrade...

Sometimes, you're surprised by the lessons you learn when you least expect them. I listened to this podcast from @davidsenra.com with Jonathan Ross expecting to learn about AI. I learned, also, so many valuable leadership lessons I was expecting. open.spotify.com/episode/1A08...

Decouple decision making from outcome quality. There's always randomness/luck involved. But, more often than not, if you have a good decision making process, good outcomes follow.

"How We Migrated a Production Cluster from Ingress NGINX to Gateway API" by Husne Ozmen www.fairwinds.com/blog/migrate...

"How to Test Infrastructure as Code" by Alex Leventer www.pulumi.com/blog/how-to-...

Something I still struggle with regularly: "no need to take note of this, I'll remember it". I rarely do. For me, note taking, even if of poor quality, is better than not note taking.

Metrics should help understand if you're achieving your mission. They should not prescribe how to achieve it.

"AI Gateway vs. Direct LLM API Integration: The Architecture Decision Defining Your AI Strategy" by Kong konghq.com/blog/enginee...

Diversity should be a priority for organizations. You shouldn't aim for people to agree blindly. In fact, it's in your own best interest if people don't Civil discourse between a diverse group usually brings the best ideas to life.

"5 anti-patterns that cause Kubernetes operator vulnerabilities" by Ugo Giordano and Rishabh Singh developers.redhat.com/articles/202...

Revision 180 is out! @david-koslicki.bsky.social #devops #sre #platformengineering open.substack.com/pub/embracer...

Your understanding of somebody else's problem will always be limited, because you'll not be in the thick of it. If you find ways to share the same problems as your customers (e.g. dogfodding) you'll build much better solutions.

"Terraform MCP server: Four real-world AI infrastructure patterns" by YuZhao Zhang www.hashicorp.com/en/blog/terr...

If you want to talk about ownership and accountability, you need to stay with something long enough to live with the consequences of your decisions.

"Blue/green deployments on Kubernetes with Argo Rollouts" by Jubril Oyetunji

"Make the easy things easy, and the hard things possible" Larry Wall Platform teams should take this to heart.