@petergyang Building my company from Amsterdam. Office is a 12 minute bike ride, kids in school down the street, dinner with friends two nights a week. Still shipping.
The tradeoff Americans assume exists mostly doesn’t.
@emollick Forecasters had 3-4 hours pegged for December. Got there in May. The curve isn’t the interesting question anymore. The interesting one: are the companies deploying AI measuring which side of it they’re on?
@garrytan The trap underneath this imo: most founders think “make the first fire” means build harder. It usually means test cheaper. You don’t need money to find out if people want your product
Excited to share our most powerful new Claude Code feature: dynamic workflows!
Mention "workflow" in a prompt and Claude will dynamically create an orchestration plan that it strictly follows, allowing you to confidently trust that every stage happens in the right order even across 100s of agents.
Claude Opus 4.8 is out today. It's our strongest coding model yet: up on SWE-bench Pro (from 64.3 to 69.2) and noticeably more honest about its own work. It tells you when it's unsure and catches its own bugs instead of declaring victory early. Same price as 4.7.
@forgebitz Both can be true with the same AI. The split is measurement imho. If you can attribute output to a team, you hire to scale where the value lands. But if you can’t, you cut because it’s the only lever you can count.
@sama Two years of “AI will reshape the labor market at historic scale” answered with a $250M foundation. Either the speeches were wrong or the budget is.
This take is everywhere now and it keeps missing the same thing.
If you couldn’t measure productivity before AI, you can’t measure AI’s ROI after. The tool didn’t break your metrics but it exposed that you never had them.
Most companies bought AI the way they bought every tool before it: top down, no baseline, no owner, no attribution. Then they act surprised when “proportional gains” don’t show up in numbers they were never tracking in the first place.
The COOs saying this out loud aren’t wrong that the ROI is unclear. They’re admitting their measurement was always unclear and AI just put a price tag on it.
Blaming the tool is easier than admitting the operating discipline was never there.
@emollick This is just FinOps again imo: “Who should get tokens and how do we control it” is word-for-word the cloud-spend panic from 2018. The answer was never exotic: meter it, attribute it to an owner, give them a budget. Tokens aren’t a new problem..
@johnawahba@tszzl Thx, good distinction! GDP works because it’s money someone else chose to give you. Token spend is money you gave the GPU.. I do think only one of those is evidence anyone valued the output
@ycombinator@twolabsai Caregiving is the right wedge for one reason: it’s the rare job where demand is rising and the workforce is vanishing at the same time. Aging societies can’t hire enough humans at any wage. It’s filling a gap that’s already empty.
True, but it compresses fast. Once everyone has AI, “humans with AI” stops being the edge. The next line is humans with AI and domain context vs humans with AI and none.
The tool gets commoditized in a year.
What doesn’t: knowing which problem is worth solving and having the taste to know when the output is good. That part still isn’t in the model imho
The aggregate pattern holds, but the comfort in it is misleading. “More jobs overall” and “the displaced person gets one of them” are different claims.
Every automation wave grew total employment and still wrecked specific people in the middle who didn’t transfer cleanly. The optimist and doomer takes are both too tidy imo. More jobs, yes. Not the same jobs, not the same people, and the gap is the whole problem.
True, and it complicates the usual story. The “US automates ruthlessly, Europe protects jobs” narrative is mostly backwards. Sweden automated the transactional work years ago and still has higher employment and stronger labor protection than most of the US. Automating the checkout didn’t gut the workforce. It moved people off the till. The iPad-instead-of-a-face thing is a design choice, not a law of automation.
The pattern nobody names: the same labs warning loudest about mass joblessness are the ones selling the tool and positioning themselves as the moral authority on the fallout. Doesn’t make the warning wrong. But “we have a duty to be honest about what we’re building” is easier to say than “we have a choice about what we build.”