Post

Follows the manufacturing triangle mantra of: "You can have a great product, at a great price, in a short amount of time - pick two of the three"

This is where, for me, working from specs & tests on typed languages has been a crucial part of it.

I think of software development in general as a triangle of Fast, Quality, Cheap

Yeah somebody else pointed that out in the thread, had a good back and forth about it

Nice triple constraint! I’ve found similar building small tools. Keeping the scope narrow seems to make AI assistance much more useful because I can still reason about the whole product and check the result properly.

The "frontend vs backend vs fullstack" discourse got completely eclipsed by AI/vibecodig, anyone can now do anything, it seems.

this pattern also happens *claps hands together and creates a small disturbance in the wind which is heard by his mac studio which writes a one-shot prompt for world peace (as a service)*

ngl the trick is having access to the servers

I think it's a "know which areas the tricky/non-obvious bugs are likely hiding out in" sense that comes from experience that leads to prompting/specifying/pushing back on that makes for a good chunk of that

I’m curious too because if it’s permanent (as I suspect, because I suspect the missing piece here is judgement), that’s very bad, because how do you train up juniors? If it’s the latter, that’s a lot less worrisome.

Cheap, fast, or good was what I was told 30 years ago.

I think it's a variety of factors. Agentic coding already incentivizes greater investment in testing rigor because it helps the agent write correct code and it gives you assurance against its mistakes. Pair that with greater scope or more ambitious projects, and now you have another reason for rigor

Like, if you soup together all code, surely the end result does certain things often. That's basically letters, learnable bits. Probably an inefficient language because AI bad at job, but that doesn't matter.

you can definitely do all three, but it comes at the expense of quite a lot of dollars

You know what sucks is, that is such a derisive statement but it IS an important part of parsing what's happening. There is a skillset to AI coding that keeps getting discounted

This.

I like blitzcreep that's great

And a lot for developers my self included are doing a *ton* of maintained related things we never really had the appetite for in the past to do due to the time and effort to do.

Right, what was formerly scarce is increasingly … not.

This also implies that modularization and encapsulation has become immensely more valuable, because it is the principle that confines scope even within extremely large projects. Then you get all three.

Way back in the day when I was in school, someone programmed an interactive graph with three vectors at 60° angles. They were labeled speed, quality, price with directions faster, better, lower respectively. If you clicked and pulled any of the vectors, the other two would shrink.

I meant where is Claude Code *itself* on this spectrum, since its makers regularly claim the whole thing is vibe coded :)

With earlier generations it was more like pick-one or pick-none if you weren’t very careful.

Everything in software engineering is you can have 2 out of 3. Unless it’s a negative. Always breaks? Runs slow? Hard to debug? I can do all 3 with ease.

I also find myself rapidly iterating on tasks and developing tests more now. I'll modify how something is done a handful of times, and write it so that it's more easily unit tested, which then leads to a larger bank of test that get written to confirm behavior now and prevent regression later.

time to slow has NEVER been faster

I do think that has a big effect. Stakeholder coordination is a way more visible drag on development pace now

I've heard that AI coding is the same old "pick 2: fast, cheap, robust" but its actually now "pick 3" and you can only pick "fast"

Scope offers more opportunities for flawed results. I'd have expected AI to slowly and carefully test in a few more defined areas before completely opening up to the public. The rash of bugs is no surprise, the rush to go public seems driven by a desperate need to show revenues to investors.