Peloton for the Mind

Mar 26th, 2026

What if “bicycle for the mind” never meant that it’s a way to speed up, but a way to exercise? Since in 2026 nobody uses bikes to exercise anymore, the more understandable metaphor would be “a Peloton for the mind,” an ergometer to get your mind-muscles into motion at all before they atrophy. Then the goal is not to be quick, to get stuff done – at least not for you, personally, – but to learn and do something interesting. The old-skool Hacker ethos. The vibe you still have in niche communities of embedded system tinkerers, of fantasy console game makers, of 6502 assembly fans, etc. etc. There’s infinite ways to have fun and learn with a computer. And there’s ways to make a living using a computer in various ways. If these two goals don’t align, you need a tool to make the chores go over as quickly as possible, so you have more time for the interesting things.

Does Bad Code in Your Context Window Misalign LLM Recommendations?

Mar 20th, 2026

Put away your Betteridge’s Law of Headlines, I genuinely don’t know the answer :) In “Training large language models on narrow tasks can lead to broad misalignment”, Betley et al found that fine-tuning LLM’s with e.g. insecure coding practices can affect the outcome of non-coding related answers in a bad way.

Make the pi Coding Agent Identify the Model in Commits

Mar 17th, 2026

In November, I posted how to make Claude Code mark itself as the author of git commits. You can do the same with OpenAI Codex.

Identify Codex CLI Git Commits for an Audit Trail

Mar 17th, 2026

Agentic engineering practices changed from “you should commit everything yourself” to giving the agentic tools more autonomy for example in order to try out different things to implement a change. Small git commits are part of that toolset. Here’s how to teach Codex CLI to self-identify as the git author.

FreeBSDKit: Swift Package to Write Capability-Aware FreeBSD Apps

Mar 11th, 2026

Shout-out to FreeBSDKit:

FreeBSDKit provides idiomatic Swift, C and C++ interfaces to FreeBSD’s unique system features including Capsicum sandboxing, jails, process descriptors, kqueue-based signal handling, and inter-process communication with descriptor passing. The framework embraces move-only semantics (~Copyable) to model resource ownership explicitly in the type system.

“FreeBSD” hooked me, “move-only semantics” got me interested, “jail” made me read the README.

I only know about FreeBSD’s features from Oliver Epper, but everytime he talk about it, I want to play with the OS. These features sound generally useful to run applications in relative isolation without the Docker overhead. Having a genuinely attractive Swift package for this makes the urge to experiment even stronger!

FreeBSD Jails as an OS-level virtualization option in particular got me interested for NAS/home server work. Code snippet for jail management:

import Jails
import Descriptors

// Build jail parameters
var iov = JailIOVector()
iov.add(key: "name", value: "myjail")
iov.add(key: "path", value: "/jail/myjail")
iov.add(key: "host.hostname", value: "jailed.local")
iov.add(key: "persist", value: true)

// Create jail and get descriptor
let flags: JailSetFlags = [.create, .getDesc, .ownDesc]
var jailDesc = try SystemJailDescriptor.set(iov: &iov, flags: flags)

// Attach current process to jail
try jailDesc.attach()

// Remove jail (requires owning descriptor)
try jailDesc.remove()

Zettelkasten Productivity: Books Written per Books Ingested

Mar 7th, 2026

Over on the Zettelkasten forums, Edmund Gröpl, in a post: “On Zettelkasten Productivity”, proposes to reframe the “how many books have I written” lifetime achievement number as “how many books have grown out of what I read”, showing how small the ‘tip’ of our productivity is compared to the huge foundation of what we read.