A nice mix this time, one reference resource, one product launch, two workflow pieces, and one architectural argument.
Ryan Chandler’s Lesser Known Laravel Glossary is exactly what it sounds like: a searchable, categorized reference of Laravel features that intermediate developers often don’t know exist. Things like Model::withoutTimestamps(), the Lottery class for probabilistic feature flags, DB::whenQueryingForLongerThan() for per-request query time monitoring, and once() for request-scoped memoization. 37 entries and counting, filterable by category and skill level. It launched right after Ryan’s Laracon EU talk and it shows, the curation is tight.
Ahmad Awais’ Command Code launch announces something genuinely different: a coding agent that learns from your actual coding behavior rather than rules you write down. Every accept, reject, and edit becomes a signal that updates a symbolic constraint layer on top of the base LLM. The result is stored in a human-readable taste.md file you can inspect and override. The rules-file comparison is sharp — rules decay because you write them once and forget to update them; taste compounds because it updates every session. The benchmarks show correction loops dropping from 4+ edits per task to under 1 after a month. I’m watching this closely.
The PrimeLine post on Claude Code Hooks is the most comprehensive practical guide I’ve found on this topic. Claude Code has 25 hook event types across the full session lifecycle, only three can block actions (PreToolUse, PermissionRequest, ConfigChange), everything else is observe-and-inject. The article walks through five production hooks with full code: a security tier check that blocks dangerous commands on protected branches, a delegation enforcer that scores prompts and routes to appropriate subagents, a context warning that tracks approximate token usage, an auto-formatter that runs after every file edit, and a pre-compact state saver. The six-months-of-production-results section at the end is worth reading on its own.
Matt Rothenberg’s OG Image Generation on the Edge is a fun architectural post. He embeds his OG image design as a hidden <template> element on each page, then has a Cloudflare Worker visit the page, extract the template, and screenshot it. Same HTML, same CSS, same fonts as the actual page, no template drift. R2 for durability, Cache API for speed, a small admin UI for cache inspection. It’s experimental and he says so, but the approach is clever and the reference implementation is on GitHub.
Simon Willison’s AI Should Help Us Produce Better Code is a necessary corrective to the “just ship it faster” interpretation of agentic development. His argument: shipping worse code with agents is a choice. Refactoring tasks that are conceptually simple but time-consuming, renaming a concept across a large codebase, splitting bloated files, removing duplicate functionality, are exactly what agents are good at. He calls the practice of capturing what worked at the end of each session and feeding it back into future agent runs “compound engineering.” That habit alone is worth building.