Written before LLMs made Big Rewrites an absolute doddle.

ameliaquining 3 hours ago | parent | next [-]

LLMs have meaningfully decreased the cost of rewrites, such that many rewrites that wouldn't previously have made sense now do. They are still significantly costly and often a bad idea.

	▲	hinkley 35 minutes ago \| parent [-]
		They also remove the secret menu item of rewrites which is to have some peace from management because you’re not delivering anything for a long time while you do the rewrite. I predict AI will cause the number of rewrites people ask for to go down. AI is better at doing Ship of Theseus rewrites too, and those are the superior option because they leave the system available for hotfixes at all times.

▲

m463 18 minutes ago | parent | prev | next [-]

it is sort of amazing what I think is practical.

"rewrite in python." "no, let's try rust."

"I don't like the grand piano in the basement, let's see how it looks upstairs"

▲

Calavar 3 hours ago | parent | prev | next [-]

An absolute doddle assuming you have an absolutely excellent test suite. Otherwise on any nontrivial codebase you will still have regressions that make it to production and you will waste your time squashing bugs rather than writing new features.

▲

paimapi 3 hours ago | parent [-]

the thing companies seem to be finding out is that quality is much more important when you're shipping 3-5x more per sprint and incident rates are multiplying at an even higher factor. allocating even less time to code review and testing is the norm, however, and it feels like the modern Agile SDLC is breaking a bit under this new mode

there's good skills to write test suites but each codebase needs its own bespoke implementation paired with so much institutional knowledge. but companies keep laying off those knowledge holders and QA to boot

it's almost like shareholders > leadership > middle managers see 'enshittification' as aspirational and are doing their best to implement a systems-level strategy in pursuit of it

▲

hinkley 28 minutes ago | parent [-]

I have a rule of thumb that goes something like, if you want to get people to use something 2x as much you need to make it 4x more reliable. The reliability becomes a enticement, and you will often underestimate how much more popular it will get, so if you aim for 2x you will likely see 3x and still have a lower incident rate per week than you experienced with lower traffic.

None of your users or managers think about errors as fraction of interactions. Even despite trying to label things as 99.99% reliable. Some managers will claim they do, but they aren’t being honest with themselves. Usually when they get mad it will be about having to deal with a problem happening twice in a week or a month. So now they’re yelling at you while you want to yell at them about being dishonest.

	▲	paimapi 16 minutes ago \| parent [-]
		right, that was always the most surprising thing I saw when I made the career leap into tech for all of it's vaunted 'big data' and 'metrics-driven outcomes' and whatnot, as soon as a single medium-to-large user starts complaining about an obscure component or wanting new features it was all hands on deck, roadmaps be damned, data out the window, who cares about CVEs/tech debt/fixing MVPs, let's do it MVP (again) instead of building something maintainable and extensible and then forget all about it like a child with ADHD when the next largeish user sends your C-suite an angry DM meanwhile the core features that most people are happy with and the entire reason they chose your solution gets ignored and one day you look in your lunchbox and the food is all gone and C-suite's golden parachuted out to some cushier job somewhere else

▲

Exoristos 44 minutes ago | parent | prev [-]

As an American English speaker, I'm going to assume "doddle" is a synonym for quagmire.