Uses — My Stack & Setup

Explicit graphs beat implicit chains — when the flow is a graph, you can see it, test it, and fix it, instead of trusting a loop you can't inspect.

The model should return a schema I can validate, not prose I have to parse — I trust a typed contract over a natural-language promise every time.

A single protocol beats a pile of bespoke adapters — the whole point is that tools plug in without my agent knowing their internals.

Environment-grounded verifiers over LLM-judge theater — check the side effect actually happened, because a model will narrate work it never did.

Primary, unapologetically — the ecosystem and the ecosystem's momentum are worth more than any language's elegance, and I'm fluent enough to see inside the math.

Occasional, and only where the front-end forces my hand — the types are welcome, the ecosystem sprawl is not.

The best thinking environment ever built for interrogating data — a notebook is a conversation, and conversations are where insight lives.

Boring in the best way — reliable, documented, and it refuses to hide the model's assumptions behind a shiny API.

Tensors you can grab and inspect beat a black box — I can see every gradient, and seeing is how I trust a model at all.

Nobody loves it, everybody needs it — and the moment you fight it instead of using it, your data is wrong anyway.

It works on my machine is an excuse, and containers are the polite way to retire that excuse forever.

Overkill until the day it isn't — I reach for it when the system gets big enough that manual babysitting costs more than the learning curve.

Parallelism without the ceremony — it turns a single-machine prototype into a distributed job without rewriting the logic I already trust.

Fast, quiet, and it sips power while a GPU box screams — for everything short of heavy training, it's simply enough.

A genuinely shallow vanity pick — I keep one because the tactile feedback makes long writing sessions feel physical instead of abstract.

Less glare, more focus — and honestly, it's the closest thing to a uniform that software work has.