How to talk to coding agents
Chapter 01of07

The beginner’s mindset

If you've been writing software with agents for a while, you might be starting to feel competent. Don't trust the feeling.

60018003000Tons more to learnYou · 1200Magnus · 2840
Most chess players sit in the 600–1000 range. Three years of practice gets you to about 1200.

AI coding is roughly three years old. Chess is five hundred. Three years of chess might get you to 1200 ELO — past most casual players, nowhere near expert. A grandmaster beats a 1200 every single game. The frontier of AI coding might be 10x as effective as your average senior engineer.

The danger of calling yourself an expert too early is that your learning stalls. You stop reading tutorials, skip the bootcamp, ignore the patterns the model enabled last week. Look at where your workflow was twelve months ago, then project that forward — most of the meta hasn't been uncovered yet.

Two responses

Sooner or later the model will let you down. It writes the wrong code. Picks the wrong abstraction. Hallucinates an API. Confidently does the opposite of what you asked. How you react in that moment shapes everything that follows.

Closed: Blame the AI.

“The model is bad.” “AI's overhyped.” “It can't do real work.”

Open: Ask what you could have done differently.

Was the prompt clear? What context was missing? What levers can you change?

Gaming culture has a useful phrase for this: skill issue. When your character keeps dying, the level isn't broken, you're missing something. It sounds flippant, but the move underneath is serious: assume, provisionally, that the bottleneck is you. It isn't always true. But this assumption helps you grow when it is.

Most of the time, the model isn't the limit. Your prompt is. Your context is. Your patience is. With the bottleneck-is-me stance, the same frustrating moment turns into a small lab: which lever would have changed the result? Each answer is a piece of edge other people aren't collecting — they've already decided the tool is bad.

Every failure is a rep

Treat each disappointment like a chess puzzle: try something, notice what actually worked, update your repertoire. One a day and the curve climbs — slowly, then all at once.

Open looptry → notice → update

One chess puzzle a day, and the curve climbs.

Closed loopblame → stop → regress

The edge you had rusts.

The beginner's mindset is the faster path. Three years in, the people pulling away are the ones still treating themselves like beginners — still reading, still experimenting, still updating when the model surprises them.

We're all beginners

Closed minds blame the model. Open minds ask what they missed and find something to learn from each transcript. Only one of those is still gaining ELO a year from now.