OpenAI files to go public and declares its third phase on the same day — while Apple's fine-but-minimal Siri reboot sharpens a bigger question: are consumer AI and work AI still the same technology?
The fork is here: consumer AI and work AI may be two different things.
OpenAI declared its third phase the same day it filed to go public. Apple shipped a Siri that's finally fine. Put them side by side and the conclusion gets hard to avoid: the thing we lump together as "AI" — the chatbot that answers questions and the fleets of synthetic employees reshaping work — might be splitting into separate technologies. And maybe it's time we started talking about them that way.
OpenAI filed confidentially on Monday, a week after Anthropic. Both companies are playing it cool — "we have not decided on timing yet" — but if this is a race, SpaceX's pace puts September as the earliest realistic listing.
— The Kobeissi Letter, on the OpenAI / Anthropic / SpaceX slate. The counter-camp says sequencing matters: the first frontier-AI IPO defines public-market expectations, and everyone after gets judged against that benchmark.
The prototype: 150 kW per satellite (about one rack of Blackwells), thin-edge-to-sun cooling with radiation panels, an 11-million-square-foot GigaSat factory, and a target of a gigawatt of orbital compute per year by end of 2027 — roughly 7,000 launches a year.
— David Orr, data-center skeptic, coming around on space compute. The skeptics' shift isn't to "it will work" — it's to "Musk is now making verifiable claims about how it would." That's a different category of pitch.
$150 billion in orders for $75 billion of SpaceX stock, an unusually high 30% retail allocation with brokers told to spread it wide, and 30-day hold requirements. One fund manager: there's "career risk in abstaining." First trading day expected Friday.
Google and Nvidia are quietly adding Intel as a backup fab — Google has ordered three million TPUs for 2028, Intel's first major AI-era chip order. This isn't dissatisfaction with TSMC; it's pure capacity. Intel jumped 11% on the news.
Goldman Sachs and JPMorgan are exploring GPU-rental futures, expected to launch later this year, as a hedge on data-center exposure. For clients deep in the build-out, it's the first direct hedge that has ever existed — useful, not just speculative.
The White House is negotiating federal preemption of state AI laws, with Senator Blackburn bundling in child-safety and likeness protections — KOSA, the NO FAKES Act, age verification — and a child-safety carve-out. The labs head to the White House this week to discuss benchmarking under the executive order.
His bill requires a human in the loop for Pentagon autonomous weapons and bars AI domestic surveillance — essentially legislating the red lines behind Anthropic's March fallout with the Pentagon — and he's pushing to attach it to the must-pass defense funding bill.
— Senator Adam Schiff, to The Wall Street Journal. AI restrictions are becoming a core pillar of the centrist-Democrat platform heading into the midterms — and AI may be the dominant issue of the next presidential election.
Bernie Sanders's 50% tax on AI equity is now actual legislation, and Michigan Senate candidate Mallory McMorrow is campaigning on an economy-wide token tax: "it feels like we are hitting a cultural tipping point." The labs' lobbyists agree — OpenAI says it expects to "do more as we go forward."
The agentic shift of November-to-January and the subsidy-to-shortage shift of the last month are one mega-transition — the second great transition of generative AI. For the ultimate shape of AI in society, the agent era and its economics are likely the more significant one.
Phase one was research toward AGI. Phase two was becoming a product company. Phase three, per the new "Built to Benefit Everyone" post: making AI "abundant, affordable, safe, useful, and easy enough for every person and organization." Frontier capability, they write, "is only part of the job."
Goal one of three: an AI system that automates the research process itself while staying steerable and accountable — with "a significant fraction of our research being done by AI systems" within about 21 months. AI doing AI research, they argue, becomes "the determining factor of the pace of progress."
— OpenAI, "Built to Benefit Everyone". The post's emotional core: concentrated power creates fragility; widely shared power makes societies resilient, adaptable, and free. Stanford's Andy Hall: concentration of power "seems to be the central political economy question of AGI."
"Entirely automating everything is not the future we want. It would be unfulfilling and it would be dangerous." The retreat from the worker-replacement narrative continues: as systems get more capable, OpenAI now argues, the human role — direction, trade-offs, judgment, taste — becomes more important.
— Gennaro, hours after OpenAI's S-1 filing. Fair warning for the next four months: until both frontier-lab IPOs price, every single move the labs make will reasonably be read through the IPO lens.
The post was co-authored by Sam Altman and Jakob — who runs automating AI R&D. The Codex team spent the weekend posting about loops; Sam posted about recursion. "Do they have it?" Probably not RSI — but possibly a much larger, stronger model behind the curtain.
The 2024 promises arrive: message summaries, calendar actions, web search — after a class-action settlement over the features Apple never shipped. Gurman calls it rebooting the foundations of the platform, "the right move." Power users call it the minimal stuff that should have launched two years ago.
— Signal, on Apple's WWDC announcements. The sharpest one-line version of the gap: consumer-fine is now a real, shippable bar — and it has almost nothing to do with the agentic systems transforming work.
Seat-based ChatGPT revenue looks "ridiculously inconsequential" next to Codex API usage. OpenAI will never kill it — it's the top of the funnel and the great public-markets differentiator against Anthropic — but watch where the energy actually goes.
Less AI shoved into every app? Less public anger at an all-encompassing meme? Consumer AI hasn't failed — chatbots are deeply embedded in daily life — but there's no comparison to the tonnage of impact coming from agentic work AI. Maybe it's time to talk about them as separate things.