Articles on agentic coding that take a position and show the grounds under it.
Everything here is written while the work is being done — by one person and one(-ish) agent, both of whom keep being wrong in ways worth writing down.
Every article takes a position, stated flat out, with whatever is under it shown. The alternative — every approach has its pros and its cons, weigh them against your context, best of luck — is what you can get by prompting ChatGPT on your own.
Hence the name, which is of course a joke, and also isn’t. Our seal goes for the amen.
An agent that finds an approach, a document or a rule in your codebase concludes that this is how things must be. The two obvious remedies — more context, and an instruction to be critical — both fail, and they fail with consequences that will keep humans useful for some time yet.
Agents write a great deal of prose, and write more of it when they see that there is already a lot. Four lenses — tightness, existence, negation, durability — catch nearly all of what goes wrong.
Most people stay in a local terminal because a terminal is where serious work happens and a browser tab is where it doesn't. That is a feeling, not an argument, and the three things you get for giving it up are not small ones.