~/how_devs_learn

You can autocomplete a career.
Or you can understand one.

A field guide for young engineers who grew up with an autocomplete that writes whole functions. The tools are incredible. The risk is handing off the one thing that made you an engineer: thinking.

measure yourself → read the field notes

./dependence --show

What it looks like

In the moment it only feels fast. The dependence shows up later, the day the tool is down and the work stops with it.

A stickman as a marionette with the Claude star pulling the strings
The marionette: every move suggested for you.
A stickman walking with the Claude star as an underarm crutch
The crutch: you get there, just not on your own.
A stickman on an IV drip fed by a Claude-star bag
The drip: a steady feed of answers on tap.

git diff your-habits

The house rules

Use AI. Just use it the way a senior engineer uses a junior dev: hand off the grunt work, then actually read what comes back. If you couldn't defend a line in review, it doesn't ship.

01 Read the error before you paste it

- paste the stack trace into a chatbot + read it first; it already told you what's wrong

02 Never merge what you can't explain

- merge the suggestion and hope it holds + merge only what you could defend in review

03 Struggle first, prompt second

- prompt the second you feel stuck + give the problem an honest 20 minutes

04 Learn the layer below

- trust the magic and move on + know what the framework is hiding

05 Type it at least once

- copy-paste a pattern you've never written + type it by hand at least once

06 Measure, then improve

- guess how dependent you are + measure it, then work the number down

ls field_notes/ | head -3

Build it to understand it

The fastest way to understand a tool is to build a small one yourself. A few worth your weekend.

// systems · databases

Build your own database from scratch

A key-value store and a relational database in Go: persistence, B-trees, and transactions, with no framework hiding the hard parts.

 // systems · networking

Build your own Redis in C/C++

Clone Redis to learn network programming, data structures, and low-level systems work from the socket up.

 // compilers

From source code to machine code

Write a compiler that turns a statically-typed language into real x64 ELF executables. The end of "compiler magic."
all field notes →