@adapalmer.bsky.social that was a great episode at @dwarkesh.bsky.social's podcast, congrats!
Galo Navarro
@varoa.net
Builds infrastructure and automation for AI workloads, Principal Eng @Midokura. Long form at http://varoa.net
AI systems, at all levels of the stack (infra and workloads, both training / inference) are complex versions of distributed systems, each with very different constraints and trade-offs www.coreweave.com/blog/why-dis...
Optimizing software for agent-legibility is quickly becoming a hard requirement... Screenshot from the last "Joy & Curiosity" issue by @thorstenball.com More on identifiers: varoa.net/2024/05/01/i...
One of the interesting parts about software engineering is that each domain you work in requires re-asessing every pattern and best practise that seemed set in stone in other domains.
I have Copilot (+ Claude) stuck in an infinite loop of refactors with small "improvements" to its own implementation of a relatively simple prompt. Feature parity with human devs is one step closer.
Listening to The Sonics' Hitch Hike, noticed that The Velvet Underground reused the riff in There She Goes, the Sonics were covering Marvin Gaye. www.youtube.com/watch?v=6-US... www.youtube.com/watch?v=KwyL... www.youtube.com/watch?v=QbW3...
Marvin Gaye Hitchhike
YouTube video by ron torres
youtube.com
Warming up to this reading list enforced backupcy thing, thanks Pocket
It takes you about the same amount of time to read this post as the amount of downtime you’d be allowed per week with 99.999% availability
“One does not replace a blown fuse until one knows just why it has blown.”
There are still 1980’s style sysadmin orgs that have survived by hiding behind the fancy labels of DevOps / SRE / Platform engineering
Platform engineering is about giving internal engineering teams the tools they need to provision, manage, understand, and operate the software they own. If you're running it for them, you're an ops team. If you're building tools, sans PM/designers, you're an infra team.
Been inspecting some pmap outputs these days, and could not find any decent visualization tools. Here is a quick one (nods to @simonwillison.net toolbox). varoa.net/tools/pmap-v...
New small tool coding workflow: open vscode, agent mode, ask for tool, go do $chore
The more we lean into vibe-based coding, the clearer it becomes that we’ll still need good operators to keep everything sane and under control
Everything associated to the “software craftsmanship” mystique is up for surprises
When I wrote about "AI will choke delivery pipelines", no-Fri-deploys was one of the practises I mentally labelled as AI-unfriendly. Interrupting the flow of code on Fri creates a glut for Mon. And bundling changes tangles latent defects, making troubleshooting harder. varoa.net/2025/04/07/a...
AI-generated code will choke delivery pipelines
Everyone is focused on the impact of AI on the production of code. Butcode isn’t just produced, it has to be consumed: built, packaged,tested, distributed, d...
varoa.net
Haha. No, @kvlly.com was (at one point) a proponent of "no Friday deploys," but I have always been anti. Very anti. As I once wrote, in my flamethrowing younger years: "Friday deploy freezes are exactly like murdering puppies." charity.wtf/2019/05/01/f...
When bundling $thing2 into $thing1 looks like a good opportunity, that's the signal for NOT doing it. If they looked like bad opportunities overengineering & overscoping would never happen.
Anybody want to come work on Tailscale's internal dev tooling/productivity, CI, bots, dashboards, etc? Pretty please? 😅
I mention the DORA report in the piece I published today. The report has a page about AI impacting sw delivery negatively. The recommendations (screenshot) about smaller changelists and robust testing make sense, but I'm somewhat skeptical that they are viable ->
Impact of Generative AI in Software Development from @dora.dev dora.dev/dora-report-... #GenAI #DevEx #GBGB
If you want a TL;DR of the post it’s probably this one!
This line is pure gold: When AI agents are able to deliver tasks based on a prompt, there won’t be enough copies of the Mythical Man Month to dissuade business folks from trying to accelerate road maps and product strategies by provisioning fleets of AI coding agents.
The type of content I expect from my timeline on a fine Monday like this
It's widely known that sharing a queue across multiple servers, rather than queue-per-server, often helps reduce latency and improve utilization. But when is one queue better? In my new blog post, I look at one case: different classes of work. Read it here: brooker.co.za/blog/2025/03...