ctrl+r #13 : Agents don't sleep, but your infra (and brain) might need to
đ§ The compute crunch is real
Been watching GitHub struggle lately. Michael Driscoll posted about their service getting shredded by automated agents running 24/7. Cloud services werenât built for this kind of constant hammering.
Tomasz Tunguz at Theory Ventures wrote about the coming AI compute crisis: real scarcity by 2026, with only the big players able to afford the best models while startups get stuck with slow service and high prices.
Makes you rethink the whole âdemocratized AIâ story.
đ ď¸ VibePod + Caveman: two agent tools worth a look
VibePod CLI: Simple way to run AI coding agents in isolated Docker containers. No setup, just vp run <agent>. Built-in analytics to compare agents side by side. The isolation model is smart: keeps your system clean while you experiment. Been wanting exactly this.
Caveman skill for Claude: Why use many token when few token do trick? Cuts 75% of tokens by making Claude talk like caveman. Removes all the pleasantries and filler, keeps technical accuracy. Brilliant for API costs when running agents at volume. (Yes, the irony of an AI talking like a caveman to save resources is not lost on me.)
đ§ Software engineer burnout because of AI
Thereâs a real paradox happening. Software engineers are scared of their jobs being taken, and at the same time excited to build things they would never have been able to build before.
But thereâs another part to the story: craving the coding skills youâve been honing for years.
I was at PyCon last week and heard the same thing from AI power users: theyâre burnt out. Our brains arenât designed to juggle 4+ agents in parallel. Youâre incredibly productive, but exhausted. And you miss the joy of writing code yourself, failing for hours on a syntax issue, then getting the lightbulb moment. Some engineers (e.g. Simon) are âdowngradingâ on purpose: learning a new language without autocomplete, writing on paper or e-ink, cutting screen time.
This resonates with some well-known figures in software engineering:
DHH on the My First Million podcast:
âthe last 3 months Iâd say has been the most churn in my mental approach to computers in the entire time Iâve been using them.â
Simon Willison on Lennyâs podcast :
âUsing coding agents well is taking every inch of my 25 years of experience as a software engineer, and it is mentally exhausting.â
Agentic coding can be a real addiction, and with any real addiction come some negative side effects.
đ ď¸ claude status in terminal
Iâve been IDE-free for ~2 months, living in Claude Code / opencode + tmux (deserves its own post). Recently I wanted better status handling to see when Claude is busy vs. waiting on my input.
I landed on two projects: opensessions (heavier, TypeScript extension) and tmux-agent-status (pure bash). Tried both, but ended up âstealingâ the best ideas and adapting them to my own setup, simpler and works better.
Hereâs what it looks like, and my dotfiles are here if you want to get inspired for your own setup!
đ What I read/watched
âYouâre right to be anxious about AI, this is how much we are buildingâ by Dumky de Wilde: the GitHub/package/paper growth charts since early 2025 are absurd. Your anxiety is rational.
âUsing coding agents well is taking every inch of my 25 years of experienceâ via Lenny Rachitsky: agent orchestration is its own skill and nobody is good at it yet.
âVibe Coding Kills Open Sourceâ: as AI gets better at generating code, contributions back to the libraries drop. Who maintains the stack when everyoneâs just prompting?
âThis guy built a $1.8B company with 2 employeesâ: telehealth shop running marketing, support, and creative through AI. The per-employee economics are getting weird.
$100M+ Advice Thatâll Piss Off Every Business Guru (ft. DHH)DHH on one of my favorite business podcasts!
Claude HowTo: visual, example-driven Claude guide with copy-paste templates. Good starting point if youâre finally getting serious with it.
PyconDE last week! Above the talks, whatâs always nice is to connect randomly with folks and share knowledge and experience.



