Cairn

Paweł Dąbrowski argues that Rails' default "Rails Way" becomes insufficient for applications exceeding roughly 100,000–150,000 lines of code. At that scale, the lack of explicit boundaries and domain modelling produces a maintenance tangle. He proposes five foundations — each building on the previous — that keep large Rails apps comprehensible and growable.

Jeremy Fairbank (Test Double) shares six practical lessons from building a Claude skill to help teams navigate complex setup documentation for a data ingestion platform. The skill acts as a "seasoned captain" — the documentation is the map, but the skill guides developers through it step-by-step, adapting to their specific codebase context. The most important insight is that context cannot be dumped wholesale into a skill. A 55,000-character document caused immediate token warnings and premature compaction. Effective skills read documents in sections, persist state across sessions, and explore the developer's existing codebase before prescribing steps.

Julia Evans migrated her websites from Tailwind CSS (used for eight years) to vanilla CSS with semantic HTML. The migration was driven by improved CSS knowledge, making Tailwind's scaffolding feel unnecessary; friction from mixing Tailwind with vanilla CSS; and a principled commitment to taking CSS seriously as a discipline. The article is partly a technical guide, partly a philosophical argument that CSS expertise deserves respect — not dismissal.

Nolan Lawson challenges the dominant narrative that AI coding tools exist to produce output as fast as possible. He argues LLMs are flexible enough to do the opposite: improve code quality through a deliberate, methodical review workflow that often takes *longer* than before. This is a different philosophy than the "10x productivity slop-cannon" approach — and he finds it more satisfying.

Johanna Larsson argues that LLM-generated content violates a fundamental implicit contract between writers and readers. The problem isn't primarily about quality or factual accuracy — it's about the *effort gap*. Historically, writing required more intellectual effort than reading; readers could reasonably trust that authors understood their subject. LLMs collapse that gap, letting authors produce volume at near-zero effort while readers still pay the full cost of attention.

The author (from Monokai) documents a two-month migration of their entire digital infrastructure from US-based services to European alternatives, driven by concerns about data sovereignty and jurisdictional control. The conclusion: running a reliable, professional digital stack, mostly on European infrastructure, is entirely feasible, and the main obstacle is inertia rather than capability.

Rails 8+ app template

A format specification for describing a visual identity to coding agents. DESIGN.md gives agents a persistent, structured understanding of a design system.

Deploy on servers, no GUI needed.

Easy self-hosting for Docker-based web apps

Using Claude and Claude Code to digest research, spike a landing page, and then apply design-driven prompting to build a product.