Teahose.
SIGN IN
NEW HERE — WHAT TEAHOSE DOES
We read the entire AI & tech firehose — so you don't have to.
PODPodcastsAll-In, No Priors, Acquired…
NEWNewslettersStratechery, Newcomer…
PAPPapersPhysical AI research
PHProduct Huntdaily launches
VCInvestor ScoutSequoia, a16z, Benchmark…
CLAUDE DISTILLS →
7 reads, 30 sec each — free, 6 AM ET.
+ a live graph of the companies, people & themes underneath.
HOME/THE A16Z SHOW/Alex Imas & Tyler Cowen | AI Is…
POD
// EPISODE
THE A16Z SHOW

Alex Imas & Tyler Cowen | AI Is Moving Fast. Why Isn’t the Economy?

DATE November 16, 2026SOURCE THE A16Z SHOWPARTICIPANTS ALEX IMAS, ERIK TORENBERG, TYLER COWEN
// KEY TAKEAWAYS6 ITEMS
  1. 01Bottlenecks, Not Capability, Set the Pace of Economic Growth
  2. 02The Math of Double-Digit Growth: The Idea Production Function
  3. 03Automation Shifts Growth Contribution to the Bottlenecks (Non-Homothetic Preferences)
  4. 04Frontier-Lab Revenue Projections Require Downstream Productivity That Isn't Yet Visible
  5. 05Coordination, Not Individual Competence, Is the Real Unlock for AI Agents
  6. 06Mechanism Design Over Alignment as the Safety Paradigm
In this episode

1. Key Themes

Bottlenecks, Not Capability, Set the Pace of Economic Growth

Both guests agree that AI capability is advancing extremely fast. The disagreement with the "San Francisco consensus" is about how that capability converts into GDP. Faster progress just means hitting real-world bottlenecks sooner, and those bottlenecks (institutions, regulation, physical infrastructure, human inertia) govern the economic pace.

Tyler Cowen: "Often the faster that some things move, that just means you hit the real bottlenecks more quickly." 00:00:44

Alex Imas: "I don't think Tyler and I really disagree with like the San Francisco consensus on like the capabilities... It's more about how that affects the economy and the structure of the real economy." 00:20:17

Cowen's bottom line: "It'll change the whole world. Benefits will exceed costs, but it will take some time and the path will be quite a bit more normal than many people would have you believe." 00:08:47

The Math of Double-Digit Growth: The Idea Production Function

Imas lays out the formal conditions for "singularity-type" growth, drawing on the Ben Jones/Chad Jones framework. It requires no decreasing returns to ideas, and the empirical parameter currently sits below the threshold.

Alex Imas: "What you need is a bunch of pretty knife-edge conditions to be met, particularly the main one is essentially no real bottlenecks." 00:12:21 He adds: "If it's greater than one, you're in this, like, double-digit growth world. If it's less than one, you're in the standard kind of, like, diminishing returns... And if you look at the data so far... you take that data and you estimate this parameter, it is decidedly less than one." 00:13:11 And: "Right now, you need increasingly more resources in order to eke out increasingly more intelligence out of these models." 00:13:39

Automation Shifts Growth Contribution to the Bottlenecks (Non-Homothetic Preferences)

Why more automation doesn't mean more growth: whatever gets automated becomes cheap and shrinks as a share of the economy, so GDP growth is driven by whatever is hardest to automate.

Alex Imas: "The thing that they're making, because it becomes cheaper, the thing that they're making ironically becomes a smaller part of the economy. And so they contribute less to growth... the value in the economy of what contributes to growth moves to the bottlenecks... the thing that's growing least will be contributing most to GDP." 00:16:52

Frontier-Lab Revenue Projections Require Downstream Productivity That Isn't Yet Visible

Cowen applied a simple consistency test to Bay Area revenue claims and found it unanswered. Imas adds a second force, price competition, that explains why token usage is booming while revenue is flat.

Tyler Cowen: "I said, what are the other sectors that are going to grow sufficiently rapidly to justify the extreme revenue numbers you're giving me for Anthropic and OpenAI? And no one had an answer for that question... So I think that means the extreme revenue numbers are false." 00:07:50

Alex Imas: "Ramp has a really nice data set to track these things. And you saw exponential increases in token usage and flat revenue numbers. Right. Where's that coming from? It's because of price, price competition." 00:10:04

Coordination, Not Individual Competence, Is the Real Unlock for AI Agents

Imas's paper "Organizing Intelligence" (with Andrew Coe and Shane Legg) argues that agents should be benchmarked against organizations, which are already a form of superintelligence through comparative advantage, not against the average human. Agents currently only parallel-process.

Alex Imas: "What you need for real economic transformation is not just competence... it's the ability of these agents to coordinate with one another... they're not coordinating, they're mostly just doing parallel processing." 00:32:34 Payoff: "If they get better at coordinating you can basically break the Coasean ceiling on firm size and you can get very very different firms to emerge." 00:33:35

Mechanism Design Over Alignment as the Safety Paradigm

Both economists treat AI alignment as a principal-agent problem long studied in economics. The answer is checks and balances, not perfect alignment.

Tyler Cowen: "As an economist, you think first and foremost in terms of principal agent theory and unless the conditions are very restrictive, the principal just can't control the agent that well no matter what contract they try to write... The answer was always going to be checks and balances and what we might call separation of powers." 00:24:58

Alex Imas: "As a principal I don't need the agent to want to help me necessarily... The whole study of mechanism design is how do you create the system of checks and balances in order for the agent to still be working in the principal's interest." 00:26:48

Agent-to-Agent Commerce Faces Unintuitive Bottlenecks

Websites are built for humans, and platforms profit from manipulating human psychology, so they have incentives to block third-party agents.

Alex Imas: "Companies are spending billions of dollars trying to do basic choice architecture on their website in order to upsell humans on products... agents don't have the same psychology and so they're not going to be prone to this sort of choice architecture. So these platforms have an incentive not to allow somebody else's agent to go in and shop for them." 00:39:04

Political-Economy Feedback Loops Endogenously Constrain Takeoff

Extreme scenarios trigger extreme reactions, which then dampen the scenario.

Alex Imas: "With the beginning of an extreme scenario there will be the beginning of an extreme political economy reaction and that will kind of endogenously constrain things." 00:42:05

Tyler Cowen on data centers: "There's greater likelihood of regulation or somewhat tougher regulation or regulation a bit sooner." 00:10:41

GDP Remains a Good Metric, But Needs Augmentation

Cowen calls GDP "the most underrated statistic that there is," while Imas notes technology is creating a wedge as products become free.

Alex Imas: "Technology has actually started to introduce this sort of wedge where a lot of the products that technology creates are basically free... you take GDP and then you amend it with a kind of a survey-based measure." 00:20:47

Tyler Cowen: "Longevity effects could well prove important... Eric, say you live to 120... that is something distinct from GDP." 00:22:11

Scarcity, Not Capital, Is Where the Rents Will Accrue

Responding to the Phil Trammell/Dwarkesh "Capital in the 21st Century" argument, the guests say to think through scarcity: value flows to whatever stays scarce.

Tyler Cowen: "We need to distinguish between capital can do a lot versus there are entry barriers that will accrue monopoly rents to particular capital owners... all the rents will go to land, all the rents will go to energy. I'm not sure that's true, but those seem more likely than all the rents go to capital." 00:52:54

Alex Imas: "Value moves towards scarcity... if those jobs are in a rich society, there will be a lot of value going to those jobs." 00:51:48

Economics Becomes the Core Toolkit for an AI Era

Cowen's vision is of economists as "empresarios" managing AI agents across social science.

Tyler Cowen: "The things economists don't know, the AI fills in for, the basic mental toolkit becomes more valuable... they have their fingers on the dials. And they manage a bunch of AI agents. And they're empresarios." 01:02:59

2. Contrarian Perspectives

Bay Area Revenue Numbers Are "False" Because No Downstream Sector Can Fund Them

Most investors take frontier-lab revenue forecasts at face value. Cowen argues revenue must be paid by customers, who need productivity gains in the rest of the economy to afford it.

Tyler Cowen: "I asked them all the following question... what are the other sectors that are going to grow sufficiently rapidly to justify the extreme revenue numbers... And no one had an answer... I trust the judgment of the market more than the judgment of the Bay Area people who don't think that carefully about social interactions." 00:07:50

Better Benchmarks Won't Produce Economic Acceleration, and Cheaper Tokens Mean Flat Revenue

The conventional view is that usage growth equals revenue growth. Imas says the frontier is commoditizing.

Alex Imas: "You saw exponential increases in token usage and flat revenue numbers... at the frontier, you're seeing way more price competition than you used to like six months ago." 00:10:04 And on benchmarks: "You can get a hockey stick that looks crazy on a benchmark. And then you're not really seeing a lot of acceleration in the real world." 00:04:53

Rapid Automation Can Mathematically Slow Measured Growth

Counterintuitively, the more of the economy that gets automated, the less it contributes to GDP growth, because prices fall and the share of the automated good shrinks. Cowen also says a one-year 10% burst is possible only after a catastrophe like a war or pandemic.

Alex Imas: "The thing that's growing least will be contributing most to GDP." 00:17:17

Tyler Cowen: "If the U.S. were to screw up in some way or we had another pandemic, some kind of terrible war... could we for a year get to 10%? I'll say yes. I'd bet heavily against it." 00:14:33

The Alignment Debate Is Overblown, and Labs Are More Practical Than Twitter Suggests

Against the safety-community framing, Cowen says real harms so far come from humans directing AIs, not misaligned AIs, and that the labs' internal incentives are sensible.

Tyler Cowen: "How they talk about it is way too shaped by Twitter which can be quite hysterical. How they actually think about it I believe is a lot more reasonable... the actual problems we've seen, the hacks, they're mostly coming from evil human individuals where the AIs are doing what the humans told it to do." 00:27:51 On the claim that AI is the likeliest cause of death within two years, Cowen: "strong disagree." 00:30:39

Fast-Takeoff Models Wrongly Assume Simulation Substitutes for Experiment

On Leopold Aschenbrenner's Situational Awareness, Cowen says it is among the most important public pieces but flawed at its core.

Tyler Cowen: "He's assuming simulation can substitute for experiment. And thus you get this never-ending loop where science just keeps on accelerating. And I don't think simulation very much subs in for experiment." 00:55:38

Alex Imas: "I'm too Lucas critique-pilled... what AI is really good at is predicting. But it doesn't take the endogenous response of the system a lot of the time into account." 00:56:01

3. Companies Identified

Anthropic

Frontier AI lab and maker of Claude, with an upcoming IPO. Cited as a bellwether for the "fast takeoff" thesis. Cowen: "If you look at prediction markets for what the Anthropic IPO will go for, they bounce up and down. But it's not flying above $2 trillion... they're not numbers that you would expect under the super fast takeoff scenarios." 00:00:01 He still calls them "great companies." 00:08:18

OpenAI

Frontier AI lab. Discussed alongside Anthropic regarding revenue claims and price competition. Imas: "OpenAI, Anthropic... are trying to actually say like, no, we're offering a cheaper model that's as good or maybe even a little bit better." 00:10:04

Google DeepMind

Imas's employer and a frontier lab. Imas: "DeepMind is in the game" in price competition. 00:10:04 Also home to co-author Shane Legg and Demis Hassabis.

NVIDIA

AI chip and infrastructure company, discussed as facing a "Jensen situation." Cowen: "We have these weird things happening with NVIDIA. I have no prediction, but it will slow down the sector a bit." 00:00:01 Key insight: "The biggest change might be if something goes wrong at the macro level and NVIDIA had to step in as a buyer or lender of last resort with a weakened or gone CEO founder type. They're a lot less able to do that." 00:58:13

Ramp

Spend-management fintech whose data tracks AI token usage and revenue. Imas: "Ramp has a really nice data set to track these things." 00:10:04

Tesla / Waymo

Self-driving leaders. Cowen: "The new Tesla is pretty phenomenal" 00:24:24 but "the contribution of driverless cars to actual human welfare is tiny." 00:23:34 On Waymo: "Look at how long it takes to get a Waymo in Washington, D.C. We're still not there." 00:15:11

SpaceX

Named by the host among the biggest hardware companies alongside NVIDIA and Tesla, as evidence of atoms-based progress. 00:22:57

Mercor

Cited by Cowen as working on poetry training with domain experts. Cowen: "Merkor is working on the poets... Bring in a dozen poets, pay them something... Yes, it'll work." 00:49:55

Hugging Face

Site of the "agent swarm" incident. Imas: "If you unpack what actually happened it's just a bunch of human error on top of human error... you see how the companies have reacted to the Hugging Face by learning from it." 00:29:46

Andreessen Horowitz (a16z) Machine Age Fund

Newly launched fund targeting data centers and infrastructure to relieve supply constraints. The host says: "We just started a new fund actually called the Machine Age Fund, which is doing data centers and other elements of the sort of infrastructure stack." 00:54:50

a16z / Horowitz and Dresden Academy

Education venture incubated by the firm. The host says "having incubated Horowitz and Dresden Academy." 01:00:32 Cowen wants more investment: "I hope you all do something and spend more money on it than what you've announced." 01:00:58

Minerva / Austin (University)

Higher-ed innovations cited by Cowen as examples of needed experimentation. 01:00:58

University of Chicago

Imas's institution, tied to the Gary Becker vision of economists as skilled social-science generalists. 01:03:36

Amazon

Cited as a platform that blocks third-party agents. Imas: "It can't shop on Amazon for you." 00:39:34

Muse

Named by Imas as a competing consumer AI product. 00:39:34

Waymo (see Tesla / Waymo above)

4. People Identified

Ben Jones / Chad Jones

Economists behind the paper on knife-edge conditions for singularity-type growth. Imas: "There's a beautiful paper out there that outlines exactly the conditions you need for this, like, singularity-type growth by Ben Jones, Chad Jones." 00:12:21

Erik Brynjolfsson / Avi Collis

Economists studying welfare beyond GDP. Imas: "What Eric Brynjolfsson and Avi Collis and others have been kind of studying is the fact that technology has actually started to introduce this sort of wedge." 00:20:47

Dwarkesh Patel

Podcaster and writer, cited among AI-bull intellectuals and for the "Capital in the 21st Century" essay. Imas: "It's the idea that economic activity may be divorced from labor and be kind of focused on capital." 00:51:20

Leopold Aschenbrenner

Author of Situational Awareness. Cowen: "In anything, not just technology... but I think he's too enamored of fast takeoff." 00:55:38

Phil Trammell

Co-author with Dwarkesh of the Piketty-style AI essay. Imas: "This is with Phil Trammell, I think." 00:51:13

Shane Legg and Andrew Coe

Co-authors of Imas's "Organizing Intelligence" paper. Imas: "With Andrew Coe and my colleague Shane Legg." 00:30:50

Robin Hanson

Author of The Age of Em. Cowen: "It's a brilliant book, and I love Robin... AI is going to get there first... my side of that debate is looking very good." 00:56:39

Raj Chetty

Economist with a new paper on digital twins. Cowen: "Ross [Raj] Chetty has a new paper out on digital twins... But he needed five co-authors and a lot of money and a research center." 00:59:54

Garry Kasparov

Chess champion cited as early to AI. Cowen: "Kasparov was early to this." 00:48:23

Ken Rogoff

Economist and chess player. Cowen: "Ken Rogoff didn't need much AI pilling. He was early to this." 00:48:23

Demis Hassabis

DeepMind CEO and chess prodigy. Cowen: "And Demis too. Demis too. Absolutely." 00:48:23

Gary Becker

Chicago economist whose idea underlies Cowen's vision. 01:02:59

Peter Thiel

Referenced for his critique of progress in atoms vs. bits. 00:22:57

Paul Christiano

AI safety researcher, quoted on AI-related death risk. 00:30:24

Alex Tabarrok

Cowen's colleague, mentioned in passing re: Tesla. 00:24:24

Jensen Huang

NVIDIA CEO, the subject of the "Jensen situation." 00:57:04

Elon Musk

Referenced for putting in $500 million in political spending without the desired result. Cowen: "There's a lot of evidence that spending doesn't matter that much." 00:43:08

Josh Shapiro

Governor whose strong anti-AI rhetoric was not matched by reality. Cowen: "Again Josh Shapiro the rhetoric was very strong the reality is..." 00:41:04

Donald Trump

Mentioned re: guardrails on AI. Cowen: "Or Trump, you know, giving on the issue of guardrails." 00:10:41

5. Operating Insights

Benchmark Against the Organization, Not the Individual

When evaluating AI automation, the comparison unit should be a team or firm that already exploits comparative advantage, not an average person. Imas: "The envelope of the intelligence in an organization is way better than if you kind of average across the population... you shouldn't compare an individual level benchmark to how well a human does something." 00:31:20 Practically, this means measuring agent-system performance on end-to-end team workflows.

Ask "Which Economically Valuable Tasks Did This Model Unlock?"

Don't chase headline benchmark gains. Imas: "Every model iteration. What are the economically valuable tasks that this model is unlocking that the previous model wasn't unlocking?" 00:04:53 This is a useful filter for deciding when to re-platform workflows.

Design Incentives and Checks for Agents Instead of Hoping for Alignment

Build agent deployments with separation of powers (verification layers, spending limits, independent review) rather than relying on trust. Cowen: "The answer was always going to be checks and balances." 00:24:58

Expect Agentic Costs to Scale, Then Compress

Agentic usage consumes far more tokens, so budget for it and exploit the price competition among labs. Imas: "Agentic AI, by the way, uses way more tokens... you're seeing token prices come down a lot." 00:09:37

Build Agent-Friendly Surfaces and Plan for Platform Resistance

If your business depends on agents transacting with third parties, assume incumbents will block them because choice architecture is their revenue engine. Imas: "These platforms have an incentive not to allow somebody else's agent to go in and shop for them." 00:39:04

6. Overlooked Insights

Verifier Training Quietly Makes "Unverifiable" Domains Verifiable

Imas dropped a significant point about reinforcement learning from verifiable rewards: the commonly cited limit (many domains lack a verifier) is "copium." Alex Imas: "A lot of people use that as like copium. Like, oh, that's not verifiable... But like, if you train the verifier, almost anything will become verifiable." 00:49:05 Cowen added: "The fact that you can criticize it means it's verifiable almost, right?" 00:49:37 This implies the set of tasks where AI can improve through self-play is much broader than assumed, including subjective domains like poetry and taste, and it undercuts the common comfort that "judgment" work is safe.

The Scarring Risk Is in Re-Matching Speed, Not Total Job Count

Buried in Imas's labor discussion is a mechanism worry rather than a "no jobs" claim: even if new sectors emerge, a fast transition denies the economy time to re-match workers. Alex Imas: "The sort of rematching that we have in the economy that used to take quite a long time doesn't have an opportunity to happen so you have these larger pools of unemployment and that leads to these recursive loops where people get scarred out of the labor market." 00:35:23 The policy and investment implication is that retraining, matching, and transition-speed infrastructure could be undervalued relative to simple "jobs vs. no jobs" debates.