E126: Big Tech blow-out, Powell's recession warning, lab-grown meat, RFK Jr shakes up race & more

Using the official arbiter of U.S. business cycles, the National Bureau of Economic Research (NBER), the last declared recession ran from February 2020 (peak) to April 2020 (trough), and NBER lists no subsequent peaks or troughs—meaning no new recession has been dated through at least late 2025. (nber.org) Real GDP growth was solidly positive in 2023 (around 2.5–3.1% for the year), indicating expansion rather than contraction. (reuters.com) Available data and commentary for 2024–2025 likewise show continued—if slower—growth and only a single quarter of mild contraction in Q1 2025, which analysts explicitly note does not constitute a recession. (desjardins.com) Because no NBER-defined recession occurred in the "near term" after April 2023 or during the 2023 economic cycle, Sacks’s prediction that the U.S. was “headed for…a recession” in that window did not come true.

Evidence from 2023 suggests that large language models had the technical potential to handle a substantial share of knowledge‑work tasks, but it does not pin down Jason’s specific claim well enough to call it clearly right or wrong.

Key points:

• OpenAI’s “GPTs are GPTs” paper (based on GPT‑4) estimated that, with access to an LLM, about 15% of all U.S. worker tasks could already be completed significantly faster at the same quality, and that when you include LLM‑powered tools, this share rises to roughly 47–56% of all tasks; around 19% of workers might see at least 50% of their tasks affected. (arxiv.org) These figures imply capabilities well in the range Jason is talking about, and often above 30% for some white‑collar roles.
• Goldman Sachs’ 2023 analysis found that generative AI could substitute up to about one‑fourth of current work tasks overall, with especially high exposure (25–50% of tasks) in many white‑collar occupations such as administrative and legal work. (gspublishing.com) That again puts a 30% task share within the plausible range of what LLMs could handle in principle.
• McKinsey’s 2023 report on the economic potential of generative AI concluded that current generative AI and related technologies had the potential to automate work activities that absorb 60–70% of employees’ time, with the biggest effects in knowledge‑intensive, language‑heavy activities (communication, documentation, supervision, etc.). (courses.cfte.education) This strongly supports the idea that a very large fraction of knowledge work tasks are technically automatable.
• Experimental evidence from MIT in mid‑2023 showed that ChatGPT substantially raised productivity for mid‑level professionals performing realistic writing tasks: time to complete tasks fell about 40%, while independent evaluators rated output quality about 18% higher, and judged it suitable for workplace use. (news.mit.edu) That supports the claim that LLM outputs can reach “practically usable” quality in real business contexts.

However:

• These studies typically measure potential or speed/quality gains on subsets of tasks, or aggregate exposure across all workers, not “what fraction of a typical white‑collar knowledge worker’s entire task mix can be done by LLMs at business‑ready quality by December 31, 2023.” Their task definitions, exposure thresholds, and populations differ, so none of them map cleanly to Jason’s “30% of knowledge work” figure.
• Some estimates focusing on LLMs alone (without full surrounding tooling) find lower immediate impact—on the order of 15% of tasks significantly sped up—only reaching 47–56% when assuming well‑designed software built on top of LLMs, which was still emerging in 2023 and unevenly deployed. (arxiv.org) That makes it unclear whether, as actually available and integrated by end‑2023, LLM‑based systems truly crossed a robust 30% threshold for the typical knowledge worker.

Given that credible 2023 research brackets the technically automatable share of tasks anywhere from about 15% (LLM alone) up to well over 50% (with supporting tools) depending on definitions and assumptions, Jason’s “30%” is plausible and directionally consistent with the literature. But because the available data don’t directly answer his specific, worker‑level claim, and the exact percentage is highly definition‑dependent, the prediction cannot be determined to be clearly correct or clearly incorrect.

Hence the verdict: ambiguous.

The prediction claimed that essentially any white‑collar/knowledge worker who did not start using modern AI tools in 2023 would be out of a job and unable to compete in the labor market by roughly 2025.

By late 2025, survey data show that:

• Only about 21% of U.S. workers say at least some of their work is done with AI; 65% say they don’t use AI much or at all. (pewresearch.org)
• A February 2025 Pew survey finds 69% of workers do not use AI chatbots at work at all, including 40% who have never used them for work and 29% who haven’t even heard of workplace chatbot use. (pewresearch.org)
• Gallup reports that even by 2025 only 27% of white‑collar employees are frequent users of AI at work, meaning the clear majority of white‑collar workers are not heavy users. (gallup.com)

At the same time, the overall U.S. unemployment rate in late 2025 is around 4.4%, not remotely suggestive of mass job loss among the majority of workers who rarely or never use AI. (reuters.com) Other research likewise shows that only about half of workers worldwide used any AI in their role in the past 12 months, with just 14% using generative AI daily—again implying that large numbers of non‑users remain employed and competitive. (pwc.com)

Because millions of white‑collar/knowledge workers still have jobs despite not using AI tools regularly (and many never having used them at all), the universal claim that such workers would be "out of a job" and unable to compete by this timeframe is contradicted by the data. Therefore, the prediction is wrong.

Across the April 2024–roughly October 2025 window Jason pointed to (“five quarter ten”), there is no evidence that the effective cost of white‑collar knowledge work broadly fell by about 90% (to 10% of its prior level) due to AI.

Evidence from firm- and task-level studies:

• A Stanford/MIT field experiment on a large customer-support operation found that giving agents a generative‑AI assistant raised productivity about 14% on average (up to ~35% for the least‑skilled), implying on the order of a ~10–25% effective labor‑cost reduction for that task, not 90%. (cnbc.com)
• A synthesis by the Penn Wharton Budget Model, summarizing multiple real‑world gen‑AI deployments, reports task‑level labor cost savings in the 10–55% range, with an average of roughly 25%, and projects those savings might rise toward 40% over coming decades—still far from a 90% reduction. (budgetmodel.wharton.upenn.edu)
• A controlled experiment with ~750 BCG consultants using GPT‑4 showed about 40% performance improvement on certain knowledge‑work tasks, again implying substantial but nowhere‑near‑10x cost improvements, and only for specific task types, not all white‑collar work. (bcg.com)

Adoption and average impact across workers:

• A Federal Reserve Bank of St. Louis study using late‑2024 survey data finds that only about 21.8% of U.S. workers used generative AI in the previous week; among those users, reported time savings averaged 5.4% of work hours, and when averaged over all workers, the gain was only 1.4% of total hours—orders of magnitude away from a 90% cost collapse. (stlouisfed.org)
• PwC’s 2025 Global Workforce survey reports that as of 2025 only 14% of workers use gen‑AI daily, with about half of workers having used AI at all in the prior 12 months, underscoring that intensive AI leverage is still limited to a minority of the workforce. (pwc.com)

Macro productivity data:

• U.S. nonfarm business labor productivity—a broad proxy for output per hour—rose 1.2% in 2023 and then around 2–3% in 2024, far from the ~900% jump in efficiency that would accompany a 90% drop in effective labor cost for most white‑collar work. (bls.gov)

Taken together, current evidence suggests:

• Generative AI has produced meaningful but incremental productivity and cost gains for many knowledge‑work tasks (often in the 10–50% range in controlled or early‑adopter settings).
• Adoption and intensity of use remain partial, so economy‑wide effects on white‑collar knowledge work costs are modest, not transformative at 10x scale.
• By the end of the roughly 1–2.5 year window from April 2023, neither micro‑level studies nor macro statistics support a ~90% reduction in the effective cost of knowledge work.

Because the prediction was both quantitatively extreme (90% drop) and time‑bound to a window that has now passed, and available data show much smaller, partial gains, the prediction is best classified as wrong.

It’s too early to definitively judge a 5–10 year holding-period prediction made in April 2023, because we are only about 2.5 years into that window.

What we can say so far:

• The San Francisco office market remains deeply distressed. Q4 2024 vacancy was still about 30.2%, even as Colliers declared the market had just reached its trough and only begun a tentative recovery. (colliers.com)
• By Q3 2025, vacancy was still around 31.8%, with availability above 35% and broker commentary that a “significant amount of space must be absorbed before a meaningful recovery can take hold.” (cresa.com)
• Mid‑2025 estimates show asking office rents down roughly 30% from their 2019 peak, and top-tier buildings still carrying vacancy rates above 25%, indicating values and cash flows have not recovered to levels that would make many distressed 2023 tower buys clearly successful yet. (mmcginvest.com)
• A San Francisco Controller’s Office analysis suggests office property values may not return to pre‑pandemic levels until after 2034, implying a decade‑plus recovery path from 2024—consistent with a multi‑year, not quick, rebound. (sfchronicle.com)
• Distressed Class A office trades at large discounts continue (e.g., a major Financial District complex sold around half the prior loan value in 2025), underscoring ongoing stress in exactly the kind of large towers the prediction referenced. (hughesmarino.com)

These data all support the spirit of Sacks’s claim that buyers of heavily vacant towers in 2023 would not see a rapid rebound, and that a long hold is likely required. However, the specific horizon of 5–10 years (i.e., recovery by roughly 2028–2033) cannot yet be confirmed or falsified, since those dates are still in the future.

Therefore, the correct classification today is “inconclusive (too early)”: the prediction is consistent with current conditions and official forecasts, but it has not yet reached the point where it can be definitively judged right or wrong.

Available data show that S.F.’s office vacancy did rise significantly above 30% after April 2023, peaking in the high‑30s, but has not (so far) clearly reached 40% on standard vacancy measures, and the “few years” window from April 2023 has not fully elapsed.

Key points:

• Around the time of the podcast, CBRE data put San Francisco’s office vacancy roughly in the low‑30% range; for example, by July 2023 the San Francisco Standard, citing CBRE, reported a record 31.6% vacancy rate and said it was expected to keep rising. (sfstandard.com)
• By late 2023, vacancy had already climbed much higher: ABC7, again citing CBRE, reported “a little over 35%” vacancy in downtown San Francisco as of December 1, 2023, the highest ever recorded at that time. (abc7news.com)
• Through 2024, CBRE’s figures show vacancy continuing to worsen into the high‑30s. Axios reported 36.7% in Q1 2024 and 36.8% in Q2 2024, both new records. (axios.com) The Real Deal, citing CBRE, reported that vacancy was projected to hit 37.0% in Q2 2024 and then “rose past 37%” to about 37.3% in Q3 2024, again setting a record. (therealdeal.com)
• Those same CBRE-based reports note that availability (vacant plus actively marketed space) reached roughly 39–39.4% in mid‑2024, i.e., near but still below 40%. (therealdeal.com)
• By 2025, multiple CBRE summaries indicate the market has started to improve rather than continue rising toward 40%: vacancy edged down to about 36.6% in Q1 2025 and then to 35.1% in Q2 2025, with three consecutive quarters of positive absorption driven largely by AI tenants. (sfchronicle.com)

Interpretation:

• Directionally, Sacks’s claim that vacancy would “get worse before it gets better” was correct: it rose from low‑30s to about 37% before beginning to decline.
• However, the more specific part of the prediction – that the real downtown vacancy could reach around 40% as leases rolled off – has not yet clearly materialized in the aggregate vacancy statistics. Vacancy peaked in the 37–38% range; availability briefly approached but did not obviously exceed 40%.
• Because he framed 40% as a plausible level “as” leases roll off over the subsequent few years, and that few‑years window from April 2023 (roughly through at least April 2026) has not yet finished, we cannot definitively say the 40% threshold will not be hit in a later downturn.

Given that (a) vacancy so far has risen substantially but only to the high‑30s, and (b) the forecast horizon has not fully run its course, the fairest assessment as of November 30, 2025 is that the prediction’s 40% figure is not yet confirmed or refuted. Hence: inconclusive (too early).

As of November 30, 2025, only about 2.5 years have passed since the April 28, 2023 prediction, and Sacks’ claim concerned a period of at least five years (into 2028) and possibly up to a decade, so it cannot yet be definitively confirmed or falsified. Current data show San Francisco’s office vacancy still extremely elevated—around 30–35% in late 2024 and early 2025, compared with pre‑pandemic “healthy” levels near 4.7–6%—so downtown has certainly not returned to a normal 5–15% vacancy range yet, which is exactly what one would expect this early in his timeline. (cresa.com) Longer‑term analyses from the city and multiple real estate research firms project that office values and vacancies may take a decade or more to get back to pre‑COVID norms, which qualitatively aligns with the spirit of his forecast but are still just projections, not outcomes. (sfchronicle.com) Because we have not yet reached the earliest point at which his prediction could be proven wrong (a sub‑15% vacancy rate before 2028), the prediction’s accuracy is still inconclusive (too early to tell).

By November 30, 2025, the U.S. has not experienced an officially recognized recession since the brief COVID downturn in early 2020. The National Bureau of Economic Research’s (NBER) business‑cycle chronology still lists the most recent peak as February 2020 and the trough as April 2020, with no subsequent peak or recession dated after that, implying the expansion that began in May 2020 is ongoing. (nber.org)

Economic data over the 2023–2025 period show slowing but continued growth rather than a broad, sustained contraction. Real GDP did contract slightly at a –0.3% annualized rate in Q1 2025, but detailed analysis notes that final domestic demand was still growing, and explicitly concludes that the U.S. economy was not in recession that quarter. (desjardins.com) Major forecasters through 2024–2025 described the outlook as a “soft landing” or at worst a mild slowdown, repeatedly emphasizing that the economy was not in recession, even as risks rose. (goldmansachs.com) Other contemporaneous commentary likewise characterizes the national economy as weak or vulnerable but not yet in recession as of late 2025. (economictimes.indiatimes.com)

Since more than two years have passed since Sacks’ April 28, 2023 statement and no NBER‑defined recession or clear broad downturn has occurred in that interval, his prediction that the U.S. was “still going to have a recession” in the near term has, to date, not come true. Thus, on the evidence available by late 2025, the prediction is best classified as wrong rather than merely “too early to tell.”

Summary of the prediction Friedberg relays a Merck executive’s claim that fetal bovine serum (FBS) sales for cellular‑meat production would:

1. Rise to about $1B/year, then
2. Drop rapidly to near zero,
3. Because cellular‑meat companies would be unable to make money at the required cost structure, causing that use case (cultivated meat using FBS) to collapse.

As of late 2025, the available evidence does not let us clearly confirm or falsify that scenario.

1. We don’t have segment‑level data for FBS used in cultivated meat Public data describe the overall FBS market (primarily research, biopharma, and vaccines), not a separate line for cultivated‑meat customers. Recent market reports put the entire global FBS market at roughly US$1B in 2024, with modest growth projected through the 2020s rather than a collapse. (businessresearchinsights.com)
Because these figures aggregate all uses, they don’t tell us whether cultivated meat specifically ever drove ~$1B/year of FBS demand or subsequently fell toward zero. That key numerical part of the prediction cannot be tested with current public data.

2. The cultivated‑meat sector has not collapsed; it is small, stressed, but still growing The global cultured‑meat market is estimated at about US$336.8M in 2024, with forecasts to exceed US$3.2B by 2033, and analysts count 170+ companies and ~20+ production sites worldwide, alongside expanding regulatory approvals in the US, Singapore, Israel, and Australia. (globenewswire.com) This indicates an industry that is still in an early, high‑uncertainty phase but not one that has clearly collapsed.

At the same time, the sector faces major headwinds:

• High production costs (often $50–100/lb vs. $4–6/lb for conventional meat) with growth media a dominant cost driver. (suscof.com)
• Sharp declines in venture funding and layoffs at leading firms like UPSIDE Foods, as well as cancelled or delayed plants. (wired.com)
• Political and regulatory pushback, including state‑level bans on cultivated meat in Florida, Alabama, Texas, and others. (apnews.com)
  These issues support the spirit of the prediction that the economics and politics are very challenging, but they do not yet demonstrate that the industry has definitively failed.

3. Companies are moving off FBS rather than simply buying $1B then going to zero Even before 2023, leading cultivated‑meat firms were already working to eliminate FBS entirely for cost and ethical reasons. For example, Mosa Meat announced in 2020 that it had removed FBS from its media and cut the cost of its animal‑free medium by 88×. (cell.ag) Since then, there has been rapid innovation in serum‑free or animal‑free growth media:

• Multus (UK) opened what it calls the first commercial‑scale facility for serum‑free growth media in early 2024. (cultivatedmeats.org)
• Simple Planet (South Korea) and others have unveiled serum‑free, food‑grade media that claim up to 99.8% cost reduction vs. conventional FBS‑containing media, with global launches planned from 2025 onward. (newtechfoods.com)
• Multiple startups (e.g., BiOM Farms) explicitly focus on animal‑free media to replace FBS, citing its cost, variability, and ethical issues. (proteinproductiontechnology.com)
• Academic work continues to develop serum‑free and reduced‑serum formulations for cultivated meat, motivated by both cost and environmental concerns. (sciencedirect.com)

This trend suggests that rather than a future in which cultivated‑meat companies collectively buy ~$1B/year of FBS and then abruptly stop because the business fails, the more likely path is progressive substitution of FBS with cheaper, serum‑free alternatives as part of an effort to make the business model work.

4. Why the prediction is “inconclusive” rather than clearly right or wrong To call the prediction right, we would need evidence that:

• FBS sales to cultivated‑meat producers rose to around $1B/year, and
• Then rapidly fell toward zero,
• Specifically because cultivated‑meat companies proved unable to build a viable business at the required cost structure.

So far:

• We lack data on FBS sales broken out by the cultivated‑meat segment, so we cannot verify the $1B‑then‑zero pattern. Available data only show that the overall FBS market is roughly $1B and still growing modestly. (businessresearchinsights.com)
• The cultivated‑meat sector is clearly under financial and political pressure, but it is still attracting investment, gaining regulatory approvals, building new facilities, and working aggressively on cost reductions, including serum‑free media. (wired.com) It has not obviously failed or disappeared yet.
• The timeframe of the prediction appears longer‑term; we are only ~2.5 years past the podcast episode, and large industry boom‑and‑bust dynamics for a new technology often play out over a decade or more.

To call it wrong already, we would need strong reasons to believe that such a boom‑and‑bust in FBS demand for cultured meat will never materialize. While current trends (rapid shift to serum‑free media and a total FBS market only around $1B across all uses) make the literal “$1B of FBS just for cell meat, then zero” scenario look unlikely, it is still speculative to rule it out definitively at this early stage.

Because:

• Key quantitative elements (the FBS revenue path in this specific use) are unmeasurable with current public data, and
• The underlying industry is still evolving and has neither clearly succeeded nor clearly failed,

the fairest assessment today is that the prediction’s outcome is not yet knowable.

Conclusion: the prediction remains inconclusive as of November 30, 2025; there is not enough observable, segment‑specific or long‑term evidence to judge it clearly right or wrong.

As of November 30, 2025, there isn’t enough elapsed time or clear outcome to judge Friedberg’s full prediction.

His claim had two parts:

1. Timing and capital needs: that recombinant protein (precision fermentation) and cultivated meat firms would need several more years beyond 2023 and several billion dollars more.
2. Ultimate outcome: that they would ultimately succeed in making these technologies work at commercially viable, cost‑competitive scales.

Status of cultivated (lab‑grown) meat (cellular agriculture)
– Multiple analyses in 2024–2025 state that cultivated meat is still years from commercial viability, with products confined to pilot runs and a few high‑end restaurants, not mass retail. Costs remain many multiples higher than conventional meat, and industry observers explicitly say commercial‑scale production is still “years away.” (sezarroverseas.com)
– Recent cost estimates put cultivated meat at roughly $25–100+ per kg versus a few dollars per kg for conventional meat, indicating it is not yet cost‑competitive. (quickmarketpitch.com)
– Companies like UPSIDE Foods have faced layoffs and have paused or downsized large‑scale plant plans, underscoring that the technology is still in a difficult, capital‑intensive scale‑up phase rather than having already “arrived.” (newtechfoods.com)
These facts are consistent with Friedberg’s “several more years” and “several billion dollars” framing, but they do not yet show whether the industry will reach true cost‑competitive scale.

Status of recombinant proteins via precision fermentation
– Precision‑fermented proteins (e.g., dairy proteins) are commercially produced but, on average, still cost significantly more than conventional animal proteins: estimates for 2024 put many precision‑fermentation proteins at $20–100+ per kg, compared to $3–5 per kg for dairy proteins. (foodwrite.co.uk)
– Market and consulting reports explicitly say precision‑fermentation proteins are not yet broadly cost‑competitive, but project cost parity in the late 2020s (around 2027–2030) if large‑scale facilities and process improvements are built—i.e., still “several more years” away. (news.sustainability-directory.com)
– Investment data show billions of dollars continuing to flow into precision fermentation and cultivated meat between 2023 and 2025, with alternative protein investments around $3.2B in 2024–2025 alone and multiple single rounds in the hundreds of millions (e.g., Perfect Day, UPSIDE, GOOD Meat), supporting his point that multi‑billion‑dollar funding is required. (quickmarketpitch.com)

Why the verdict is “inconclusive”
– Only about 2½ years have passed since April 2023; Friedberg spoke of “several more years,” which most reasonably implies a longer horizon (roughly the later 2020s or beyond).
– The core falsifiable part—that these technologies will ultimately succeed in achieving commercially viable, cost‑competitive scale—concerns an outcome that industry analyses themselves place mostly in the future (late 2020s or 2030s). (mckinsey.com)
– Current evidence strongly supports his diagnosis that the field would remain expensive and capital‑hungry for years, but it does not yet confirm or refute his long‑term success claim.

Because the decisive part of the prediction (“we are going to get there”) refers to a longer time frame than has elapsed and the technologies have not yet either clearly succeeded or definitively failed at cost‑competitive scale, the correct classification as of November 2025 is “inconclusive (too early).”

The prediction is explicitly long-term and open-ended — Friedberg says recombinant/synthetic-biology-based production will eventually become ubiquitous and displace traditional animal-sourced production for all major classes of proteins and cell types. As of November 30, 2025, this has clearly not happened yet (traditional animal agriculture still overwhelmingly dominates global protein supply; precision fermentation and cultivated meat remain niche and expensive), but the claim is about an unspecified future endpoint, not about being achieved by 2025.

Because:

• Recombinant and precision‑fermentation proteins are growing (e.g., rennet, certain dairy proteins, specialty ingredients), but they are not close to being ubiquitous across all major protein and cell-type categories.
• Cultivated (lab‑grown) meat is still pre‑mass‑market, with high costs and regulatory and scaling challenges, not a displacement of traditional animal production.

the only thing we can say is that the prediction has not yet come true, but it also cannot yet be judged definitively wrong, since its time horizon is unspecified and clearly beyond the present.

Therefore, the correct status is "inconclusive" (too early to tell) rather than right or wrong.