🇺🇦 Sviatoslav Sydorenko
Alyssa CoghlanOh, it just clicked: LLM hype is what you get when you stack outsourcing hype on top of no code hype and tech silver bullet hype. So it's not one hype cycle, it's multiple hype cycles running in parallel, and it may not be clear which of them any given proponent is afflicted by.
Huh, had a fascinating conversation today that made me realise the agentic AI cost & risk conversation looks *very* different when you consider it from the point of view of an organisation that *already* outsources a lot of their internal software development to third party suppliers, as it's a matter of taking the project management & QA processes you already wrap around your outsourcing vendors, and wrapping those around an agentic AI backed development process instead.
It's essentially "outsourcing as a service", replacing what can be spectacularly complex contract negotiations with a comparatively straightforward price per token. Even if prices per token were to increase to actually profitable/sustainable levels, a single displaced outsourcing contract will pay for a *lot* of tokens.
This changes where I see employment risks from agentic AI development:
* likely devastating for software contracting shops (why negotiate a contract for developers of unknown skill levels vs subscribing to a token based service?)
* largely negative for software shops that avoid outsourcing (it's easier to scale token usage up and down than it is to go through hiring/lay-off cycles or negotiate outsourcing contracts)
* potentially neutral for non-software orgs that already outsource heavily
Appending a further related observation that came up in another thread: streamlining outsourcing of custom software development, or even routinely outsourced knowledge work in general, still isn't a big enough market to justify the current sky high evaluations and overblown rhetoric around generative AI, so this "outsourcing-as-a-service" realisation is just a glimpse of what a post-bubble world might look like, not a claim of anything more significant than that.
JP@ancoghlan Accenture-as-a-Service
Blackoverflow@ancoghlan
This assumes that the expected quality of both strategies is comparable.
Given that the experienced quality with outsourcing is often bad, that assumption may hold...
@blackoverflow Yeah, that second half of your thought was a big part of my realisation, too. If someone's primary experience with custom software development is being perpetually disappointed by external contractors, the coding LLM vendors have a much lower quality bar to reach in order to make their services commercially competitive. "Better than a sketchy body shop" is a far more plausible claim than "better than the best software engineers you have ever employed".
@blackoverflow It doesn't make the ethical and social problems with LLMs go away, but a lot of the code covered by outsourcing contracts is already protected more by trade secret law than by copyright law, so an org's IP lawyers would likely be OK with that side of things, and the insider risks from misbehaving bots are only arguably worse than those posed by external contractors, so I can see why the risk management folks would sign off as well. It all makes the paradigm shift claims plausible.
alys@ancoghlan the double cicada brood of tech hype.
Not a doctor. Or Jesus.I think this is what my geology prof calls #resonance
When seismic waves match up with a building's natural waves, the building can be destroyed. That's the science behind various architectural ways to buil earthquake resistant buildings.
These hype cycles synching up probably works the same way.
Hugo Slabbert ⚠️@ancoghlan so, three hype cycles in a trenchcoat?
Ölbaum@ancoghlan You nailed it.