The Changelog: Software Development, Open Source · Changelog Media

Kaizen! Let it crash (Friends)

January 17, 2026·1 hr 41 min·5 clips

Gerhard explains why Erlang's 'let it crash' philosophy beats defensive coding for resilient systems.

The irritation starts with CI. GitHub Actions gets credit for being everywhere, then takes the hit for builds that crawl. That turns a partner read into a real builder complaint. Namespace is the counterexample, with a pitch that stays plain: cache dependencies, Docker layers, and build artifacts so developers are not stuck waiting through coffee-length feedback loops. Then the map comes out. US East, US West, Europe, and Asia carry the obvious load, while Australia, Africa, and South America stay quieter in the background. They pick SJC. That small choice gives the segment its developer-show feel, because the work runs through flyctl, flags, prompts, and a very specific Varnish stats command. The counters are a lot. Origins, backends, memory pool, disk pool, and log counters keep piling up until even a curious technical person needs help with the wall of output. So AI shows up as a practical tool. No magic act, no keynote shine, just the thing you reach for when the terminal coughs up more signal than one pass can hold. Jared starts with Claude. His ranking is dry and telling: Claude first, Grok second, ChatGPT third, with Opus getting the first serious look. Abacus.ai gets name-checked too. The appeal comes from use, not branding: he says he has used it for a long time, pays $10 per month, and can choose models there. The mood stays loose. Even with infrastructure, regions, and Varnish internals on the table, the friends keep it warm, teasing, and curious instead of turning clinical. Kaizen is the loop. By the end, the joke is also the method: pick the rabbit hole, follow it, admit the work is never finished, and come back next time.

As heard by us

A loose, technical systems episode about slow builds, Fly instances, Varnish counters, and the awkward usefulness of AI-assisted debugging.

Kaizen! Let it crash is at its best when it treats operational mess as something worth poking at in public. The conversation opens with the familiar complaint that GitHub Actions can feel painfully slow, then moves into a live look at Fly instances, regional traffic, an SJC…

Read the full review in PlayNext →

Why you'd press play

You want faster builds, fewer waits, and a straight talk-show detour through GitHub Actions and Namespace.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from The Changelog: Software Development, Open Source