Stehfyn

It's a reductive inverse corollary, but highly skilled Blackjack players are known to hesitate hitting on 18

Piggybacking here; I'd describe it like a fish ladder. Instead of "teach" I'd say "orient." LLMs are a force whose magnitude is undeniable and increasing, but it's up to us humans to provide the theory that exerts the magnetic forces to naturally encourage them in the right directions.

Personally, this is an experience I thought about first before writing my comment. I think in the days pre-AI coding assists, I believe you describe the intrinsically human experience that's requisite to write code by hand. The wonder, the joy, the frustration, the confusion, the elation--the discovery. These days, the things I wonder about lie deeper and deeper behind more and more lines of code, through journey's that provide less and less joy, and thusly becoming more and more unreachable as I'm human, bound by an excess of things in addition to time. AI has helped me rediscover some of this sporadic creativity demonstrably due its ability to prototype recreational ideas on a whim

Professionally, I'm employed writing safety-critical avionics software. Superfluous amounts of cogent tooling putting guardrails on agents has enabled me to spend heaps more time to think deeply about how the software should work at a systemic level. The code by definition must be heavily criticized and battle-tested before it can go out the door to begin with. Albeit a beautiful part of coding, those sporadic bursts of creativity drive the code leaving my desk less and less, and I feel strongly that has made its quality paradoxically better since I'd spent much more time on broader implications and interactions.

I use a multimodal approach to defining my spec: different layers of criteria for how the software looks, behaves, what it produces, and under what constraints.

For the literal code:

• A healthy cocktail of /WX + /Wall, plus clang-tidy with very few suppressions

• An extremely opinionated mix of clang-format and LLM-generated bespoke formatting that AST-based tools cant express

• Hungarian notation; all stack locals pre-hoisted, declared in order of appearance, and separated from subsequent assignments

• Enforced dataflow: all memory accesses are bounded independent of branch resolution, with only data-oblivious indexing

• Functions have a single point of return

In a C89 workflow, this pushes agents to produce code where wrong business/domain decisions are unmistakably obvious, while eliminating the vast majority of bug classes before I ever read it.

So yeah, Ill reassert 80%, if not more.

Just because I am capable of "writing all that code", doesn't make the option preferable to defining a vast majority of spec up front and having an LLM generate an implementation. I am already going to spend the brain power on reviewing the code. I am already going to spend the brain power on pontificating edge cases, external module interactions, and next steps. Why not fast forward to that point and save 80% of the time (and brain power/attention/motivation to boot)?