random memes }

Nine of ten gone

We are about to lose nine-tenths of the folk in "software engineering".

Playing with LLMs, the code generated is not different from what I expect from most of my peers. Sure, I can read through the code, and spot both low and higher level errors. Not good enough for the code that I write, but ... not different for the bulk of my coworkers.

This is going to be ugly.

We have a whole lot of folk writing slightly-crap code. Nothing new, just the best we could do. Now the "AI" domain has reached the level of slightly-crap.

Upper management folk are going to do what they always do - jump on a bandwagon, and cover up the later disaster.

Some will jump early, and fail badly. Management barely knows how to stumble through the current model, and for the most part has no clue as to what is next. So they will fire nine of ten, then find they kept the wrong tenth. Will try to use their failure as evidence that they did not fail. Put differently, when the model changes to lose nine of ten developers - we must also lose nine of ten managers.

A few will jump carefully, and skinny-down slowly. Most engineering management fails. A smaller set learns as things change, and better serve their company. Those folk know that slower well-understood change will better serve.

In the end, folk like me will do fine. Going to take a decade, as does any large change. I am "retired", so maybe I do not care.

But the mess in-between matters. In the short-term, management train-wrecks mean a lot of folk out of work. Hiring pipelines optimized to average programmers are clueless for selecting that smaller fraction. Lots of churn, before things settle.

Some chance AI will replace folk like me. Current LLMs work off training sets. The training on the web is 90% "C", 10% "B" and 1% "A". Current LLMs are pretty good at "C". From experience, every large advance takes a decade. So figure a decade to get to "B", and another to "A".

Between needing a better algorithm, and a 1% training size, that last step might take a while.

But the next decade or so is going to be brutal.