journalwrite.html

Write

Published document

It is fair to say that using LLMs — usually referred to as artificial intelligence — is mainly a matter of delegation. Many people nowadays rely on these AI tools in their personal and professional lives. While I do finds its ubiquity to be a grave danger to society — its tendency to make mistakes and hallucinate is well known, especially in cheaper tiers with low-to-none reasoning effort — I don’t have much to say that hasn’t been said already.

Delegation is important. Experienced managers know that. It’s just as important to know what should be delegated. In software engineering, AI has slowly gone from being a smart autocomplete, to planning and executing very small changes, to being able to plan and execute a complicated feature change from start to finish decently.

Even with all that, mental fatigue doesn’t seem to be getting any better. In fact, anecdotally speaking, it seems to be deteriorating. If you want something done right, whether by hand or by an AI, you need to know what you’re doing. The more complicated the software you’re building, the harder it will be for an AI to figure out all the intricacies of the changes needed. On top of that, it doesn’t help when the generated code itself is over-complicated, hard to follow, and “unnatural,” leading to added fatigue when you need to review it.

Of course, there’s another way. You can choose to offload some of that fatigue onto a colleague. Vibe-code the whole thing, have Claude check its own work, and have it then generate several hard-to-read paragraphs for its excuse of a PR description, and ask them to review. This is wrong, naturally, but an option for sure.

My point is, you can choose to delegate everything to Claude or ChatGPT. All of your planning, writing, thinking, your life even. I fear, though, that this comes with a cost, in a use-it-or-lose-it kind of way. Critical thinking is a mental muscle, and like any muscle its strength will weaken over time, as you choose to exercise it less and less.

So maybe write. You write. Write that RFC yourself, or that PR description, or that blog post. Hone your writing skills, exercise your critical thinking, and be kind to your readers. You will have a better understanding of your work, and your readers will thank you for not writing AI slop that’s hard to comprehend. I can ask an AI any time and cut out the middleman. I want to know what you think instead.

← Back to the source
mainj / k navigateUTF-8no build step

Find a thought.

Click a result to open · Enter searches all posts

Discover more from Lazy Neurons

Subscribe now to keep reading and get access to the full archive.

Continue reading