Yesterday I Learned

2026-05-02

Maps, Margins
& Making Taste

Five things learned on May 2nd — where cycle time really lives, what responsive images finally got right, how to build codebases agents can read, and why designed habits beat guilty ones.

"

"Map your value stream. Literally follow a feature from idea to production. Write down every step. Write down how long each step takes. Write down how long things sit between steps. The gap between steps is where your cycle time lives. This will be depressing. Do it anyway. Bring snacks."

01
Delivery

The Gap Between Steps

We talk about coding speed but the time-killers live in the handoffs. A PR that sits for three days. A review queue. An approval that needs a Slack ping. These are the value stream gaps.

Value stream mapping — following a feature from idea to production, logging every step and every wait — is new vocabulary worth adding. Most people think they know the map. They don't.

There's also a secondary lesson in the reading: the mix of confirmation bias (relief that these problems exist elsewhere) and genuine learning. Both have value, but knowing which is which is the honest part.

02
Web Platform

Auto-Size Me

The RICG folks shipped sizes="auto" for lazy-loaded images. After years of writing pixel-math sizes attributes that were close-enough-but-wrong, the browser can just figure it out.

Mat's Piccalilli post contextualized the full arc — the RICG workaround, the native spec, the srcset / sizes dance. That historical context matters for understanding what actually changed versus what just got easier.

Follow-up

Audit dandenney.com-astro for where sizes="auto" could simplify existing image markup.

03
AI Craft

Teaching Machines to Read Your Room

Two nuances on CLAUDE.md: the system prompt is already ~50 lines before any project or user instructions load. What you add layers on top of meaningful existing context — the strategy is less "give Claude instructions" and more "supplement what it already knows about itself."

The AI Hero approach to agent-friendly codebases reframes the question entirely. Not just "how do I use AI to help me code?" but "how do I build the codebase so the AI can navigate it reliably?" Conventions, docs, architecture decisions — all of it becomes infrastructure for the agent, not just for humans.

04
Practice

Designed Habits Beat Guilty Ones

Raj Nandan's taste-building loop is concrete where most advice is abstract:

01

Pick one high-leverage artifact from the week — a paragraph, a label, a customer email, a key slide.

02

Generate 10–20 versions with a model.

03

For each version, write one sentence starting with "fails because..."

The step most likely to be skipped. Don't skip it.

04

Rewrite the strongest version with a hard constraint: no buzzwords, one idea per sentence, must acknowledge a real trade-off.

05

Ship it somewhere real. Observe what happens.

Also this

Morning gaming (~1 hr before work) has neuroscience behind it: mood priming, dopamine regulation, transition ritual. Not a guilty pleasure — a designed habit.

Design Notes

YIL 2026-05-02 — Maps, Margins & Making Taste

A day of practical reckoning — where cycle time really hides, what responsive images finally got right, how agents read your room, and why designed habits beat guilty ones.

accent-1 / text-amber-600 — delivery
accent-2 / text-sky-600 — platform
accent-3 / text-violet-600 — ai craft
accent-4 / text-rose-600 — practice

This is a warm, editorial light page — the visual equivalent of a well-curated morning newsletter. White background with colored section markers that feel like tabbed dividers: each learning cluster gets its own accent that appears in the category pill, section heading underline, code chips, and card borders. The value stream quote gets a full-bleed callout panel in amber-tinted cream, making it the emotional anchor before the content sections begin. Typography is confident — large hero headline, generous spacing, clean gray text hierarchy. The page should feel like opening something curated and intentional, not scrolling a feed.