This is correct. It will require non-market forces to regulate soft-landings for humans. We may see a wave of "job-preserving" legislation in the coming years but these will eventually be washed away in favor of taxing the AI economy.
115 karma · joined July 16, 2025
(I find there's a growing push-back against being fed AI-anything, so when this is suspected it seems like it generates outsized reactions)
- Did you build your own or are you farming out to say Opencode? - If you built your own, did you roll from scratch or use a framework? Any comments either way on this? - How "agentic" (or constrained as the case may be) are your agents in terms of the tools you've provided them?
[0]: https://docs.anthropic.com/en/docs/build-with-claude/prompt-...
What a silly premise. Markets don't care. All markets do is express the collective opinion; in the short term as a voting machine, in the long term as a weighing machine.
Seeing a real uptick of socio-policital prognostication from extremely smart, soaked-in-AI, tech people (like you Salvatore!), casting heavy doom-laden gestures towards the future. You're not even wrong! But this "I see something you all clearly don't" narrative, wafer thin on real analysis, packed with "the feels", coated with what-ifs.. it's sloppy thinking and I hold you to a higher standard antirez.
> Inference APIs aren’t subsidised
This is hard to pin down. There are plenty of metal companies providing hosted inference at market rates (i.e. assumed profitably if heading towards some commodity price floor). The premise that every single one of these companies is operating at a loss is unlikely. The open question is about the "off-book" training costs for the models running on these servers: are your unit economics positive when factoring training costs. And if those training costs are truly off-book, it's not a meritless argument to say the model providers are "subsidizing" the inference industry. But it's not a clear cut argument either.
Anthropic and OpenAI are their own beasts. Are their unit economics negative? Depends on the time frame you're considering. In the mid-longer run, they're staking everything on "most decidedly not negative". But what are the rest of us paying on the day OpenAI posts 50% operating margins?
Yup.
> I ... review each one
Yup.
These two practices are core to your success. GenAI hangs reliably hangs itself given longer rope.
There's a correction coming, but not the one Daniel's fearmongering about.
The current state of LLM-driven development is already several steps down the path of an end-game where the overwhelming majority of code is written by the machine; our entire HCI for "building" is going to be so far different to how we do it now that we'll look back at the "hand-rolling code era" in a similar way to how we view programming by punch-cards today. The failure modes, the "but it SUCKS for my domain", the "it's a slot machine" etc etc are not-even-wrong. They're intermediate states except where they're not.
The exceptions to this end-game will be legion and exist only to prove the end-game rule.
At 100 dev shop size you're likely to have plenty of junior and middling devs, for whom tools like CC will act as a net negative in the short-mid term (mostly by slowing down your top devs who have to shovel the shit that CC pushes out at pace and that junior/mids can't or don't catch). Your top devs (likely somewhere around 1/5 of your workforce) will deliver 80% of the benefit of something like CC.
We're not hiring junior or even early-mid devs since around Mar/Apr. These days they cost $200/mo + $X in API spend. There's a shift in the mind-work of how "dev" is being approached. It's.. alarming, but it's happening.
The painting/photography metaphor stretches way too far imo - photography was fundamentally a new output format, a new medium, an entirely new process. Agentic coding isn't that.
Or just `$ playwright`. Skip the MCP ceremony (and wasted tokens) and just have CC use CLI tools. Works remarkably well.
Honestly, it's just this. "Claude the bar button on foo modal is broken with a failed splork". And CC hunts down foo.ts, traces that it's an API call to query.ts, pulls in the associated linked model, traces the api/slork.go and will as often as not end up with "I've found the issue!" and fix it. On a one sentence prompt. I think it's called an "Oh fuck" moment the first time you see this work. And it works remarkably reliably. [handwave caveats, stupid llms, etc]
Yes, with the caveat: only on the first/zeroth shot. But even when they keep most/all of the code in context if you vibe code without incredibly strict structuring/guardrails, by the time you are 3-4 shots in, the model has "forgotten" the original arch, is duplicating data structures for what it needs _this_ shot and will gleefully end up with amnesiac-level repetitions, duplicate code that does "mostly the same" thing, all of which acts as further poison for progress. The deeper you go without human intervention the worse this gets.
You can go the other way, and it really does work. Setup strict types, clear patterns, clear structures. And intervene to explain + direct. The type of things senior engineers push back on in junior PRs. "Why didn't you just extend this existing data structure and factor that call into the trivially obvious extension of XYZ??".
"You're absolutely right!" etc.