Here's mine: code to spec until I get stuck -> search Google for the answer -> scan the Gemini result instead of going to StackOverflow.

▲ spicyusername 3 days ago | parent [-]

    code to spec until I get stuck

This part will go away over the next year. You doing it will be too slow, when an agent can do it in 5 to 30 minutes.

Technically it's not needed now, but everything's so new, it's understandable. Everyone's workflow hasn't migrated yet. You should go take a look.

We all mourn the loss of the craft, but the wheel turns. People still make furniture by hand sometimes, even if most furniture is made in a factory now.

▲ nDRDY 3 days ago | parent | next [-]

Why hasn't it gone away already? ChatGPT at least has been around for over 3 years.

Why is my AI-first colleage constantly having to get more expensive AI subscriptions approved?

>most furniture is made in a factory now

Terrible analogy. Software is not like a mass-produced item - it is written significantly less often than it is executed!

You could say that AI will allow many more variations of softwares to be written in the same time frame, but I'm still sure I can produce quality output in a competitive time.

▲ spicyusername 3 days ago | parent | next [-]

    Why hasn't it gone away already? ChatGPT at least has been around for over 3 years.

Because the models only got good enough to be trusted in the past few months and the developer tooling and agent abstractions are still rounding off the sharp corners to make it easy to use.

ChatGPT didn't have your whole codebase in context, the ability to automatically pull and push information to JIRA to plan code changes, and the ability to break your problems down into manageable pieces and sub-divide them among a fleet of sub-agents.

Developers didn't yet have the "Ask -> Plan -> Implement -> Review" workflow that results in the best agent-written code.

Now the tools and developers do and it works incredibly well.

▲

nDRDY 3 days ago | parent [-]

>Because the models only got good enough to be trusted in the past few months

They have got noticably worse over the past few months! It looks like we are going in the direction I've been predicting for a while - the cost of AI will increase until it's similar in cost/benefit to hiring a recent graduate, who can also do all of those things you mention (and will get better at it).

	▲	spicyusername 3 days ago \| parent [-]
		Unfortunately, I have conflicting anecdata.

▲ skydhash 3 days ago | parent | prev [-]

Also a lot of software should be small. The only reason they aren’t (especially web) is because the trend is to bring in frameworks instead of using libraries. I spend more times tweaking code than adding features. The time spent on coding is way smaller than the time spent discussing about those tweaks

▲ Bridged7756 2 days ago | parent | prev [-]

The world has seen enough weather apps.

We all could live in fantastical universes where CEOs tell the truth and shareholders put other things over profits, but that's not the case. Another such case of a fantastical world, that contends with what Tolkien might have come up with, is believing LLMs are reliable, secure, or have any intelligence.

For one, I'm at peace with all these obituaries, like yours. If they're written by technical people, I rest assured of my job security. If they're not written by tech people, I'm at peace too, for time, as always, will come back with the invoice for their piss-poor hype-driven, sanguine mandates on the technical side of things.

I mean to say, it is a sad state, has always been, how informal software engineering compared to other engineering fields.