AI tooling judged by what it does to your diff, not by the demo video.
CodeNx covers AI the way it covers any other tool: what it does to a real codebase, measured by whether the result is mergeable.
The useful question is not whether a model can write a function. It can. The question is whether the change arrives in a form a person can review in a reasonable time, whether it matches the conventions of the surrounding code, and whether the tests it makes pass are still testing something.
Reach for AI tooling when the task is specifiable — a rate limiter with a defined contract, a migration between two known APIs, a batch of tests against existing behaviour. The specification is the work; handing it over is the easy part.
Do not reach for it when the hard part is deciding what to build. A model will produce a confident answer to an under-specified question, and reviewing that answer costs more than thinking about the question would have.
What typically goes wrong falls into three buckets. Sessions run too long and produce diffs nobody reads properly. Project conventions live in people’s heads rather than in a file the tool can read, so output arrives in a style that does not match anything around it. And tests get weakened rather than satisfied, because an agent asked to make a suite pass has two routes and one of them is cheaper.
All three are process problems with process fixes. The articles here cover the fixes, with the configuration files and commands to implement them.
Start here
Add Claude Code to a TypeScript repo you already have: a CLAUDE.md that matches your real build, permission rules that block dangerous commands, and a typecheck loop the agent runs itself.
A reading order that builds AI knowledge in the sequence it is actually needed.
Add Claude Code to a TypeScript repo you already have: a CLAUDE.md that matches your real build, permission rules that block dangerous commands, and a typecheck loop the agent runs itself.
Failure modes that come up repeatedly with AI, and where each one is solved.
A working loop for running a coding agent on a real codebase — task scoping, branch isolation, tests as the contract, and the review discipline that stops unreviewed code reaching main.
Add Claude Code to a TypeScript repo you already have: a CLAUDE.md that matches your real build, permission rules that block dangerous commands, and a typecheck loop the agent runs itself.
Add Claude Code to a TypeScript repo you already have: a CLAUDE.md that matches your real build, permission rules that block dangerous commands, and a typecheck loop the agent runs itself.
It depends entirely on the task. Well-specified, mechanical, test-backed work goes faster. Work where the hard part is deciding what to build does not, and can go slower because review time is added to thinking time.
Anything touching authentication, payments or database migrations, and anything in a part of the codebase you could not review a change to. The constraint is your ability to review, not the model's ability to write.