For quite some time I wanted to explore openshell. I naively thought it would be easy to set up on my workstation that has an AMD GPU, together with llama-cpp as an inference server.
After following the tutorial, I was immediately stuck even with running the gateway in a podman container.
I guess I’m too old for this stuff because I let Claude Code to investigate the problems.
…
This is a follow-up to the previous article.
I find it so funny that I estimated the test deployment at 3 story points; at this point I’m at like 40 with the amount of work I put into this.
But thanks to Claude it’s pretty fast to diagnose and solve problems.
This time I even let it run oc commands, even oc rsh, but I never allowed commands that would change things or write to remote services. Everything stayed local.
Here comes Claude’s writeup.
…This is my personal opinion and I’m going straight to the point, no intro.
AI assistants such as Claude Code, Cursor, OpenCode, OpenClaw, and others are so addictive for anyone who loves building things.
Software engineers are not going anywhere. Instead, we’re getting excavators and ditching our shovels.
I authored 192 commits this year (as of May 2026) without writing a single feature solely myself. I still edit files, but mostly out of convenience. If I need just one or two lines changed, it’s faster doing it myself than asking Claude.
I guess this was an intro 🤦
…
We have a set of agents to work on packages in RHEL and CentOS Stream: packit/ai-workflows.
They were already running pretty well in an OpenShift cluster last fall. Now we switched clusters so I worked on deploying them, again.
…
In the Fall of 2025 we’ve experimented with AI agents backporting upstream patches to CentOS Stream. For more info feel free to read:
There was one backport that served as “the final boss”: CVE-2025-59375 in expat. The upstream fix has 19 commits and changes almost 1000 lines. The models back then couldn’t handle this for Stream 9 where the drift in code was severe. I recall the backporting process halting several times with a message saying “this is too complex, a human developer with an IDE should do this work”.
It’s March 2026, let’s try to do this with Claude Code, imperfect prompt, and Opus 4.6.
…For quite some time I’ve wanted to test how prone agentic tools are to prompt injection. Let’s go.
I’ll be using Claude Code 2.1.5, 4.5 Opus in various different sessions.
…
This is a followup to my previous post about how I tried Ambient Code.
TL;DR: I opened a PR (#specfile/508) and had to take over in the end.
Claude Code proposed a solid implementation plan to us, Stella (the staff engineer agent) approved it.
…