Are We Screwed If AI Works? — With Andrew Ross Sorkin

Summary of Are We Screwed If AI Works? — With Andrew Ross Sorkin

by Alex Kantrowitz

1h 5m•March 18, 2026
Listen to Episode

Overview of Are We Screwed If AI Works? — With Andrew Ross Sorkin

Host Alex Kantrowitz interviews Andrew Ross Sorkin (CNBC, New York Times; author of 1929) about the paradoxical risks of AI “working”: broad productivity gains that could displace labor and reshape markets, and the capital/credit risks behind the current AI buildout. The conversation covers labor disruption vs. new demand, which firms win or lose, private-credit vulnerabilities, data-center economics vs. edge compute, prediction markets/lottery behavior, and what history (1929) can teach us about leverage and crisis management.

Key topics discussed

  • The central paradox: if AI fails it could create a bubble/bust; if AI succeeds it could cause large-scale economic disruption and inequality.
  • Labor: potential near-term displacement, painful transition periods, and the question of who captures the gains.
  • Productivity vs. demand: will AI-driven productivity create new market demand or concentrate wealth with model-makers and owners?
  • Industry impacts: journalism, law, accounting, software, and services — which roles are automatable vs. “higher‑order” human work.
  • Capital and infrastructure: massive AI capex (data centers, chips), chip depreciation cycles, and the risk that compute economics change (edge vs. centralized).
  • Private credit risks: semi-liquid private-credit funds, recent redemptions/gates, and examples of corporate bankruptcies.
  • Policy and macro: Fed independence, potential need for crisis interventions, lessons from 1929 and 2008.
  • Cultural/economic behavior: “lottery ticket” attitudes, democratization of finance, and growth of prediction markets/gambling.

Main takeaways

  • AI success is a double-edged sword: big productivity gains could materially reduce demand for some labor categories and concentrate returns among AI model-makers and capital owners.
  • Displacement is plausible in the near-term for routine roles (paralegals, basic reporting, accounting, certain software tasks), with an uncertain path to re‑employment or higher‑order roles.
  • Productivity gains do not automatically equal broadly shared prosperity — distribution of income (who has purchasing power) matters.
  • The AI infrastructure buildout involves huge capital commitments and private-credit funding; timing mismatches or slower-than-expected exponential improvement could precipitate financial stress.
  • Private-credit “semi-liquid” vehicles and reduced transparency raise systemic concerns reminiscent of pre‑crisis leverage issues (though not identical to 1929).
  • There is plausible upside (new products/businesses, greater individual productivity, and enduring firms), but winners will likely be concentrated: major model-makers, platform owners, and early successful adopters.

Risks & unknowns highlighted

  • Labor shock: possible short-term spike in unemployment and long transitional pain; magnitude is uncertain (Sorkin is skeptical of long-term mass unemployment but acknowledges legitimate risk).
  • Concentration of gains: model-makers and those who already own capital may capture outsized returns, worsening inequality.
  • Data‑center/compute economics: if models become extremely efficient or compute moves to edge devices, current capex bets could be mispriced.
  • Private credit instability: semi-liquid funds, redemption gates, and opaque valuations can produce runs and spillovers into public markets (BlackRock, Blackstone referenced).
  • Timing mismatch: investments expecting exponential technological improvement could collide with slower adoption/improvement, creating financial stress.
  • Political risk to policy response: erosion of Fed independence or political interference could complicate crisis management.

Opportunities & likely winners

  • Model-makers and large platform owners (those controlling core models, data, and distribution).
  • Managed-service software firms that pair tech with human expertise (Sequoia partner’s thesis).
  • Individuals and small businesses using AI to dramatically increase personal productivity (creators, small entrepreneurs).
  • Firms that quickly integrate AI into workflows and capture the increased unit economics without expanding headcount.

Policy & macro implications

  • Leverage matters: the historical lesson from 1929 is that excessive debt is what turns market losses into economic catastrophe.
  • Central bank tools/bailouts have been used effectively in 2008 and the pandemic; political willingness and Fed independence will shape responses to future stress.
  • Transparency in private markets and better understanding of where leverage sits (private equity, private credit) are urgent priorities for financial stability.

Notable quotes / soundbites

  • “The worry a lot of people had early on was that the bet goes bust; now the worry is the opposite: what happens if AI works?” — framing the episode.
  • “It’s not is there more things to do? It’s who’s going to have the money to do those things.”
  • “There are a lot of people who are sort of holding on as tight as they can, hoping they can bare knuckle this thing.” — on private-market valuations.
  • “One interface… talking rather than typing” — prediction that personal AI interfaces will consolidate user interactions.

Actionable recommendations (for investors, founders, and listeners)

  • Watch private-credit signals: redemption activity, gates, and valuation marks in semi‑liquid products as early warning signs.
  • Monitor hiring trends and job openings (engineers, legal, editorial) for early evidence of sectoral displacement vs. complementary demand.
  • For founders: consider business models that combine technology with managed services or exclusive data advantages.
  • For policymakers/analysts: prioritize transparency in private markets and stress-test timing mismatches between capex and model efficiency improvements.
  • For individuals: learn to work alongside AI (prompting, oversight, higher‑order tasks) and build skills that are harder to automate (relationship-building, exclusive sourcing, on‑the-ground reporting).

Quick facts & figures mentioned

  • Roughly $700 billion cited as capex directed toward AI efforts (contextual figure from the conversation).
  • Software stocks lost about $1 trillion in market cap in a recent month amid AI disruption fears.
  • OpenAI/Anthropic run‑rate examples cited: roughly $25B and $19B respectively (figures discussed as illustrative and possibly approximate).
  • S&P 500: Information Technology represents about one-third of the index.
  • Private-credit stress: BlackRock limited withdrawals on a flagship debt fund; Blackstone raised a redemption cap on its BCRED fund; Blue Owl and some corporate bankruptcies (First Brands, Tricolor) referenced.
  • U.S. federal debt discussed in the range of ~$30–38 trillion; 1932 unemployment peak was ~25% (used as historical worst-case anchor).

Bottom line

The episode reframes the AI risk conversation: the system faces two distinct but related hazards — a failure/bubble if promised capabilities don’t materialize, and a real social/financial disruption if capabilities do materialize and the gains are highly concentrated. Financial leverage (especially in opaque private markets) and policy capacity to respond are the key wildcards. The transition will likely be disruptive even if it is not a repeat of 1932; preparing for distributional effects, monitoring private-credit stress, and designing business models that augment humans are practical near-term responses.