reading / week 12

the ratio is usually like this

read four things this week about ai tooling

one validated what already works

one taught something that was already figured out

one actually changed a workflow

one was just a good primer

the ratio tells you something


boris cherny talking about how claude code gets built

three terminals

three contexts

one product

felt less like learning

more like recognition

Building Claude Code with Boris Cherny →


mihail eric on a new era of learning software engineering

juniors starting with ai from day one

the framing is solid

but after building a governance layer because agents were lying to each other

it mostly reads like catching up

From Writing Code to Managing Agents →


the figma piece was different

design updates spacing

code doesn't know

drift just compounds

now it's one prompt to sync

not incremental

removes an entire class of bug

From Figma to Claude Code and back →


ben thompson on agents over bubbles

not a hot take

just clean structure

who's building what

why the agent layer matters more than the model layer

Agents Over Bubbles →


pattern across all four

most ai content is still framing the problem

the value shows up in artifacts that skip that

and just solve the workflow

figma mcp doesn't explain why syncing matters

it just syncs


useful filter

did it change something tomorrow

or describe something from last week

most weeks it's 1 out of 4

the one carries the rest