OpenAI's $150B conversion, Meta's AR glasses, Blue-collar boom, Risk of nuclear war

OpenAI did publicly announce and ship a formal agents product well within 12 months of September 27, 2024.

• On March 11, 2025, OpenAI published “New tools for building agents,” introducing the Responses API (with built‑in tools like web search, file search, and computer use) explicitly framed as “building blocks that will help developers and enterprises build useful and reliable agents,” and defining agents as systems that independently accomplish tasks on behalf of users. The post states that the Responses API is available to all developers starting today, i.e., a shipped product rather than a research demo. (openai.com)
• The same day, TechCrunch reported that “OpenAI launched new tools designed to help developers and enterprises build AI agents,” describing the Responses API and the open‑source Agents SDK as OpenAI’s new platform for creating production agents that can search the web, navigate sites, and work with company data. (techcrunch.com)
• OpenAI’s later AgentKit and “API Agents” documentation explicitly build on the March 2025 Responses API and Agents SDK as a unified platform for designing, deploying, and optimizing agents (Agent Builder, Agents SDK, ChatKit, etc.), reinforcing that these earlier releases were a formal agents product line, not just experiments. (openai.com)
• Separately, on January 23, 2025, OpenAI launched Operator, a browser‑controlling AI agent, as a research preview for paying Pro users, further confirming that OpenAI had moved from mere demos to a user‑facing agent product within the window, even if branded as “research preview.” (openai.com)

Because OpenAI’s Responses API + Agents SDK (and associated tooling) constitute a clearly labeled, widely available agents platform announced and shipped about 5½ months after September 27, 2024—well inside the ~12‑month horizon—the prediction that OpenAI would “publicly announce and ship a formal ‘agents’ product” in that timeframe is fulfilled.

The prediction has a 5–10 year horizon starting from September 27, 2024, i.e., it is specifically about market conditions sometime between 2029 and 2034. As of today (late 2025), we are only ~1 year into that window, so it is too early to know whether “most enterprises” will abandon legacy-priced systems of record.

What we can see so far:

• There is active technical work and industry interest in AI/agent-based enterprise systems (e.g., agentic ERP and generalist enterprise agents), but these are still framed as emerging approaches or roadmaps, not yet as mainstream replacements for core CRM/ERP/HRIS across most enterprises. (arxiv.org)
• Some high‑profile cases, like Klarna reportedly decommissioning around 1,200 SaaS apps including Salesforce in favor of an internal AI platform, show that early adopters can justify replacing traditional SaaS with AI systems, but this is one company, not evidence about the majority of enterprises. (medium.com)
• Analyst and industry pieces project that by the early‑to‑mid 2030s, AI‑driven “services‑as‑software” or RaaS/agent models may erode traditional license‑based SaaS, but these are forward‑looking forecasts, not outcomes already realized. (horsesforsources.com)
• Current pricing and deployment patterns show that traditional enterprise systems of record (CRM/ERP, etc.) remain widely used and expensive, even as AI features are added on top and some vendors begin experimenting with consumption or agent‑interaction pricing rather than strictly per‑seat licenses. (cqlsys.com)

Because the prediction is explicitly about what will happen over 5–10 years and we are far from the end of that period, and because current evidence shows only early movement rather than a decisive, broad shift where “most enterprises” can no longer justify legacy pricing, the correctness of Chamath’s prediction cannot yet be determined.

As of November 30, 2025, we are only about 14 months past the prediction date of September 27, 2024. Jason’s prediction explicitly referred to outcomes over the subsequent 2–3 years (i.e., roughly late 2026 to late 2027) for revenue-per-employee at Google, Uber, Airbnb, and Meta. Because that time window has not yet arrived, we cannot yet evaluate whether each company has in fact doubled its revenue per employee relative to the 2023–2024 baseline. Any current financial or headcount data would only show partial progress toward a 2026–2027 target, not whether the prediction ultimately holds. Therefore, it is too early to judge the prediction’s accuracy.

It is too early to determine whether this prediction is right or wrong.

The claim is explicitly about the world roughly 10 years after September 27, 2024 (around late 2034). Today’s date is November 30, 2025, so only a little over one year has passed. The scenario the predictor is talking about (late 2034) has not occurred yet, so we cannot judge outcomes that depend on that future date.

What we can say about the present is that smartphones are still the dominant personal computing interface:

• In 2025, ~59–64% of global web traffic comes from mobile devices (primarily smartphones), with desktop around 36–40%. (mynewitguys.com)
• Articles summarizing 2025 data repeatedly describe smartphones as the main way people access the internet, indicating they remain the central personal computing device. (techgaged.com)
• Major tech leaders (e.g., Sundar Pichai at Google and Andrew Bosworth at Meta) say smartphones will stay dominant for at least the next several years, even as AR/smart glasses and ambient computing evolve. (timesofindia.indiatimes.com)

However, these observations only tell us that the transition has not happened yet in 2025; they do not prove that it won’t happen by ~2034. Because the prediction’s target date is still almost nine years away, the correct evaluation status is "inconclusive" (too early) rather than "right" or "wrong."

The prediction concerns the state of personal computing and interface paradigms around ten years after the episode date of September 27, 2024, i.e., roughly September 2034. Today is November 30, 2025, so fewer than two years have passed since the prediction was made and about nine years remain before the stated horizon. It is therefore too early to determine whether traditional smartphones will have largely disappeared as the dominant personal computing device or whether browser-based interaction will have largely ceased to be primary by that time. Because the forecast is explicitly about a future state well beyond the current date, its correctness cannot yet be evaluated.

As of November 30, 2025, the preconditions for Jason’s prediction have not been met, and the time horizon (“once these things weigh a lot less…”, implicitly “over the next several years”) is far from exhausted.

Smart‑glasses are clearly progressing but are still a niche, early‑adopter product rather than a mature, mass‑market replacement for a large share of smartphone usage. Meta’s Ray‑Ban line has sold only a few million units since launch and, while successful within its niche, is described as a specialty gadget whose U.S. smart‑glasses market remains small, despite sales roughly tripling in 2025. (reuters.com) Surveys likewise show modest ownership (around mid‑single‑digit percentages) and only partial consumer interest, not anything close to universal usage. (spglobal.com) That level of adoption is far below what would be needed for glasses to handle “about one‑third” of everyday phone tasks for the average person.

On the hardware side, major players still struggle to deliver lightweight, all‑day AR glasses with phone‑class performance and battery life. Apple, for example, has repeatedly postponed or canceled AR‑glasses projects because current technology cannot yet deliver a comfortable, glasses‑sized device with sufficient power and thermal characteristics, and is instead focusing on headsets or future smart‑glasses concepts. (macworld.com) Industry analyses also highlight unresolved constraints around weight (roughly 50 g as a practical limit), battery life, and full AR capability. (ainvest.com)

Since (1) the envisioned hardware (“once these things weigh a lot less”) is not yet fully realized, (2) mainstream behavior has not shifted so that glasses routinely take over ~⅓ of smartphone tasks, and (3) the prediction’s timeframe is multi‑year and has not elapsed, it is too early to judge definitively whether Jason’s forecast will prove right or wrong.

Evidence through November 30, 2025 shows that Apple Vision Pro has remained a niche, short‑lived product rather than a widely adopted, enduring platform.

Lack of mainstream traction

• IDC estimates that Apple shipped only about 400,000 Vision Pros in 2024, versus 5.6 million Meta headsets, and notes that premium headsets over $1,000 (including Vision Pro) account for just 5–6% of the market and are largely bought by enterprises and enthusiasts, not the mass consumer market. (theverge.com)
• Business and market reports repeatedly describe Vision Pro’s reception as lukewarm and the device as a niche product: for example, Jefferies cited a “lukewarm reception” when discussing Apple’s outlook, and broader AR/VR headset shipments are projected to decline ~12% in 2025, reinforcing that this category has not gone mainstream. (investing.com)
• An August 2025 report notes that Vision Pro is “struggling to gain traction,” with fewer than 1 million units sold in the U.S. after roughly a year and a half, despite OS and app improvements. (benzinga.com) These numbers are far from the scale associated with a mainstream Apple platform (hundreds of millions of users).

Brief relevance and signs of fading prominence

• By April 2024 Apple had already cut shipment forecasts for Vision Pro due to cooling demand in its core U.S. market, signaling that initial hype did not translate into sustained broad interest. (technewsday.com)
• In 2025, multiple reports indicate Apple has scaled back or paused development of major follow‑on headsets: Reuters, summarizing Bloomberg reporting, says Apple halted work on a next‑generation Vision Pro and a cheaper N100 model to shift resources toward AI smart glasses, citing “declining momentum” and limited mainstream content. (reuters.com) MacRumors likewise reports that development of both the lighter “Vision Air” and a redesigned Vision Pro has been paused, with only an incremental M5 refresh planned, and explicitly notes the product is still unlikely to enjoy mass appeal. (macrumors.com)
• Analyst roadmaps for the M5 refresh forecast only 150,000–200,000 units shipped in 2025 and describe Vision Pro as likely to “remain a niche product,” essentially a stopgap to maintain minimal market presence rather than a growing mass platform. (macrumors.com) Meanwhile, French‑language coverage notes that production of the original M2 model was stopped in 2025 due to disappointing sales, further underscoring its short commercial life. (fr.wikipedia.org)

Synthesis

• By late 2025, Vision Pro has not achieved mainstream adoption by any reasonable metric (shipments, user base, or app ecosystem scale) and is widely characterized as niche and struggling to gain traction.
• Apple’s strategic pivot away from expanding the Vision Pro headset line and toward smart glasses, combined with very modest projected shipments and the end of production for the initial model, supports the view that Vision Pro’s window of prominence has been relatively brief and is waning, rather than expanding into a dominant, long‑lived platform.

Given this evidence, Sack’s normalized prediction—that Vision Pro would fail to gain sustained mainstream traction and would fade rather than become a broadly adopted, durable platform—is substantially borne out by the data available as of November 30, 2025.

As of November 30, 2025, only about 14 months have passed since the September 27, 2024 episode. That is too little time to judge "near to medium term" commercial success or long‑run dominance 25–30 years out.

Early evidence does suggest Meta's Ray‑Ban smart glasses are gaining traction: Meta and EssilorLuxottica report that the Ray‑Ban Meta glasses have been top‑selling products in many Ray‑Ban stores across EMEA and have driven significant sales growth, leading the companies to extend their partnership. (techcrunch.com) Independent reporting and industry analysis also indicate more than 2 million units sold by mid‑2025 with shipments roughly tripling year‑over‑year, framing the line as an early commercial success in a still‑nascent category. (ai-daily.news)

However, "commercially successful in the near to medium term" remains a forward‑looking claim, and the bolder part of the prediction—that the eventual iconic "killer" AI hardware device 25–30 years from now will not resemble today's phones, glasses, pins, etc.—is about technology adoption in the 2050s, which cannot yet be assessed. Since both elements hinge on future outcomes, the prediction cannot currently be judged fully right or wrong.

As of November 30, 2025, the prediction splits into two parts with different levels of support.

1. "AirPods‑style earbuds will evolve into effective hearing aids" – largely correct

   • In September 2024, the U.S. FDA authorized Apple’s Hearing Aid Feature software, allowing AirPods Pro 2 to function as over‑the‑counter (OTC) hearing aids for adults with mild to moderate hearing loss. The FDA notes that in a clinical study, users of the self‑fitting AirPods setup achieved similar perceived benefit and amplification performance as professionally fitted hearing aids, and the feature is explicitly regulated as an OTC hearing‑aid device. (fda.gov)
   • Multiple reports in 2025 describe AirPods Pro 2 as “clinical grade” hearing aids once the software is enabled, and regulators in countries like the UK and Australia have also cleared them as medical devices for mild‑to‑moderate hearing loss. (thetimes.co.uk)
   • Coverage emphasizes that these AirPods‑based aids can meaningfully improve speech understanding and provide benefits comparable to traditional devices for the indicated population. (opb.org)
     Conclusion: The technical/medical part of the prediction—that AirPods‑style earbuds would become genuinely effective hearing aids—has clearly materialized.

2. "…and will become socially acceptable to wear almost continuously (effectively 24/7)" – not yet realized, and constrained

   • Battery and design limitations make continuous or all‑day wear impractical. Typical hearing aids are designed for 12–16+ hours per charge, whereas AirPods Pro 2 used as hearing aids get on the order of 4–6 hours before needing to go back in the case, a gap explicitly noted in consumer and tech reviews that recommend them for occasional or situational use rather than full‑time wear. (cnbc.com)
   • Experts commenting on the new AirPods hearing‑aid capabilities in markets like Australia and in U.S. coverage emphasize that they are “not ideal for all‑day wear” and best suited to situations like restaurants or TV watching, not round‑the‑clock use. (theguardian.com)
   • Health guidance continues to warn against wearing earbuds all day because of ear‑wax build‑up, infection risk, and potential hearing damage from long exposures—again at odds with normalized 24/7 use. (gadget-faqs.com)
   • Socially, while AirPods‑like devices clearly reduce the stigma of visible hearing aids and are framed as a way to make hearing support more acceptable, etiquette commentary still notes that keeping AirPods in during conversations tends to be read as rude rather than neutral, indicating norms have not shifted to “24/7 is fine” across everyday interactions. (opb.org)
   • Adoption data are still early: only a fraction of adults with hearing loss use any hearing aids at all, and reporting describes AirPods hearing‑aid features primarily as an entry point or situational assistive tech, not as a widely adopted, full‑time replacement for traditional hearing aids. (opb.org)
     Conclusion: The specific claim that these devices would become socially acceptable to wear almost continuously (effectively 24/7) has not come true yet and is actively limited by battery life, comfort, health concerns, and remaining etiquette norms.

3. Why the overall verdict is “ambiguous” rather than “right” or “wrong”

   • The core technological prediction (earbuds becoming real hearing aids) is clearly correct.
   • The social/behavioral prediction (24/7 socially acceptable continuous wear) is clearly not our present reality, but the prediction did not include a concrete time horizon, and social norms around wearables can evolve over many years.
   • Because one major component is already correct while the other is both unfulfilled and open‑ended in timing—and because “social acceptability” is inherently hard to measure definitively—it’s not accurate to call the entire prediction simply “right” or “wrong” at this point.

On balance, the evidence supports a mixed outcome: technically right, socially unfulfilled. Given the lack of a clear deadline and the normative nature of the second part, the fairest single label today is "ambiguous."

As of November 30, 2025, less than two years have passed since the prediction was made and roughly 19 years remain until the ~2044 horizon implied by “in 20 years.” The claim is specifically about what will predominate as the dominant form of human–computer interaction for the next generation (his children’s generation) around that time, not about the situation in the mid‑2020s.

Currently, traditional direct control methods—keyboards, touchscreens, mice, and manual UI navigation—still dominate mainstream computing on phones, PCs, and tablets, even though ambient and AI‑mediated interactions (voice assistants, context‑aware services, generative‑AI copilots, etc.) are growing. There is no way to definitively assess now whether these ambient interactions will become the dominant mode by ~2044.

Because the prediction is explicitly about a 20‑year future state and that date has not yet arrived, its accuracy cannot yet be judged.

The prediction is framed over the “coming years,” which implies a multi‑year structural shift rather than something that should be fully visible by late 2025, only about a year after the podcast.

There are early signs consistent with stronger demand for in‑person human‑service work:

• Recent U.S. data show that health care and social assistance plus leisure and hospitality accounted for more than 100% of net job gains over the latest year, with other sectors stagnating or shrinking, indicating rising relative demand for face‑to‑face services. (axios.com)
• BLS long‑term projections to 2033–2034 expect particularly fast growth in human‑service‑heavy categories such as home health and personal care aides (+20.7%), personal care and service occupations overall (+6.4%), and construction and extraction (+5.6%), all above the average for all occupations. (blog.dol.gov)

However, the specific claim is that as AI and automation advance, there will be a “burgeoning and blossoming” of salaries, availability, and demand for in‑person human services. On that causal and magnitude question, the evidence is not yet clear:

• Global AI labor analyses (e.g., PwC’s 2025 AI Jobs Barometer) find the strongest wage and productivity gains in AI‑exposed sectors like finance and ICT, while less AI‑exposed sectors such as hospitality and some personal services show lower productivity gains and are not (yet) singled out for exceptional wage acceleration. (pwc.com)
• BLS projections attribute much of the growth in care and service jobs to demographics (aging population, chronic conditions) rather than to AI‑driven substitution elsewhere, and we do not yet have several years of realized wage data isolating the AI effect in trades, hospitality, fitness instruction, tutoring, and similar roles. (bls.gov)

Because (1) the stated horizon is longer than the time that has elapsed, and (2) the available data do not yet clearly show a broad, AI‑driven surge in both pay and job counts across the full range of in‑person human‑service roles he described, it is too early to say whether the prediction is ultimately right or wrong. Hence, inconclusive (too early).

The prediction has two parts:

1. Short‑term escalation into Lebanon (by the time the show aired)
   This did occur. On 23–30 September 2024, Israel launched Operation Northern Arrows, carrying out more than 1,700 airstrikes across Lebanon—over a 125% increase in Israeli strikes compared with August—killing hundreds and displacing large numbers of civilians. This marked a sharp expansion from earlier, more limited cross‑border fire. (en.wikipedia.org) On 30 September/1 October 2024 Israel then began a ground invasion of southern Lebanon, formally described as the 2024 Israeli invasion of Lebanon, confirming that the incursion had indeed “gotten bigger” in the days around airing. (downloads.regulations.gov) So the near‑term escalation call was accurate.

2. Probability (>30–40%) of a full‑blown multinational Middle East war before the November 2024 US election
   Here the specific scenario he highlighted did not happen:

   • The Lebanon front evolved into a large Israel–Hezbollah war plus an Israeli invasion of Lebanon, but contemporaneous descriptions frame it as the 2024 Israeli invasion of Lebanon and the Israel–Hezbollah conflict—not as a multi‑state regional war. (en.wikipedia.org)
   • Although Iran did launch direct missile/drone attacks on Israel in 2024 and multiple Iran‑aligned militias (in Iraq, Syria, Yemen) were active, mainstream assessments consistently spoke of the risk of a broader regional war rather than a realized multi‑state war, and noted that these conflicts had not escalated into wider regional wars by late 2024. (streetinsider.com)
   • Instead of widening into a multinational war before the US election (5 November 2024), the Israel–Hezbollah war was halted shortly afterward by a US‑ and France‑mediated Israel–Lebanon ceasefire agreement signed 26 November and effective 27 November 2024, explicitly aimed at ending hostilities between Israel and Hezbollah/Lebanon. (en.wikipedia.org) No Arab state (e.g., Egypt, Jordan, Saudi Arabia) or Iran entered into open conventional war with Israel in that period, nor did the US.

Because the core forecasted outcome—“a full blown multinational war in the Middle East” triggered by the Israel–Lebanon front before the 2024 US election—did not occur, the prediction is judged wrong for scoring purposes, even though one component (rapid escalation of Israel’s Lebanon operation between recording and airing) was correct and his stated probability (30–40%) was not a categorical claim of certainty.

The prediction tied a clear time window to Ukraine’s supposed military collapse: the speaker said Ukraine was “getting close to being combat incapable” and could collapse “in the next month… two months… six months” from the Sept 27, 2024 episode—i.e., by roughly late March 2025.

Open-source assessments of the front as of March 2025 show a severely strained but still functioning Ukrainian military, not a force that had become combat‑incapable or collapsed:

• A March 2025 campaign review notes that Russia was making slow, incremental territorial gains (about 73 square miles in the month) but that the overall strategic picture “remained largely unchanged,” with intense fighting and Ukrainian defenses still holding on multiple axes—indicating continued organized resistance, not collapse. (globalsecurity.org-www.globalsecurity.org)
• March 27–28 operational summaries describe Ukrainian Defense Forces conducting coordinated strikes that destroyed Russian armored vehicles, radar, ammunition depots, and command/logistics hubs, and inflicting large daily Russian losses in personnel and equipment—evidence that Ukraine retained the ability to plan and execute combined-arms operations. (amazing-ukraine.com)
• On March 25, 2025, Ukraine carried out a cross‑border HIMARS strike in Russia’s Belgorod region, reportedly destroying four Russian helicopters and operating on Russian territory—again inconsistent with a force that had become combat‑incapable. (theguardian.com)
• New large‑scale engagements beginning in late March and February 2025 (e.g., the Novopavlivka offensive and the Sumy offensive) explicitly list multiple Ukrainian brigades and corps as active combatants, showing that Ukrainian forces remained organized and field‑capable well beyond the six‑month horizon. (fr.wikipedia.org)

Looking beyond the prediction window, by late 2025 Ukraine’s armed forces are still clearly functioning: major battles and offensives on multiple fronts continue, and Western and Ukrainian officials are planning around Ukraine maintaining a strong post‑war army, not around a collapsed military. (reuters.com)

Because Ukraine did not become “combat‑incapable” and did not “eventually… collapse” within roughly 1–6 months after the Sept 27, 2024 episode, the prediction is best classified as wrong.

Evidence since the podcast (27 Sep 2024) broadly matches both parts of the prediction:

1. Ukraine’s military position deteriorated toward potential collapse.

   • By November 2024, major outlets were explicitly warning that parts of the Ukrainian frontline were at risk of collapse as Russian forces advanced “village after village” amid severe Ukrainian manpower and equipment shortages.(news.sky.com)
   • Analyses in 2024–25 repeatedly noted that delays in Western aid and ammunition put Ukrainian lines "at risk of collapse" and highlighted Ukraine’s serious manpower problems.(hudson.org)(en.wikipedia.org)
   • In 2025 Russia mounted a large spring and then summer offensive, capturing places such as Kurakhove (announced by Russia in January 2025) and pushing toward key hubs like Dobropillia and Pokrovsk, with Ukrainian officers describing the situation as "very difficult."(en.wikipedia.org)(en.wikipedia.org)(en.wikipedia.org)
   • Russia also opened a new front in Sumy Oblast, seizing Kostiantynivka in June 2025 and bringing Russian forces within 20–25 km of Sumy city, further straining Ukraine’s defenses.(en.wikipedia.org)
     Collectively, this supports the claim that Ukraine’s military situation continued to worsen and was widely discussed in terms of possible local or broader front-line collapse.

2. Western political/strategic pressure to intervene more directly increased as the situation worsened.

   • Even before the podcast, France’s President Emmanuel Macron had broken a taboo in February 2024 by saying that sending Western ground troops to Ukraine could not be ruled out, explicitly arguing that “nothing should be excluded” to prevent a Russian victory.(reuters.com)(kyivpost.com)
   • As Russian advances and Ukrainian shortages mounted, this debate evolved into concrete planning for Western troop deployments inside Ukraine (even if framed as non-frontline). In 2025 the UK and France began leading a “Coalition of the Willing” to prepare a multinational reassurance force to be deployed in Ukraine after a ceasefire, with around 30 countries involved in operational talks about troop contributions.(en.wikipedia.org)(euronews.com)
   • European and UK leaders repeatedly discussed this force as part of security guarantees needed because of ongoing Russian offensives and Ukraine’s vulnerability, and Zelensky himself urged coalition members to establish a framework for deploying such a reassurance force under a U.S.-backed peace plan.(reuters.com)(theguardian.com)
   • At the same time, there was strong resistance from many NATO members to full-scale combat deployments, and official positions continued to rule out NATO combat troops. But the fact that serious planning for multinational troops in Ukraine became a live issue at all – after being unthinkable early in the war – shows that political and strategic pressure to consider more direct involvement did increase as Ukraine’s position worsened.(theguardian.com)(globalsecurity.org-www.globalsecurity.org)

Because (a) Ukraine’s military position clearly deteriorated and was widely described as being at risk of collapse in parts of the front, and (b) Western leaders moved from flat rejection of any troops toward openly debating and planning for deployments of multinational forces on Ukrainian soil (even if limited and post‑ceasefire), the core directional claim of the prediction is borne out. The prediction did not say that the West would actually send combat forces, only that the temptation and pressure to intervene directly would grow as Ukraine approached collapse, which has happened in a qualified but real way.