The world is too loud. Read what matters.

The a16z Show

AI Is Accelerating. Why Isn't the Economy? The Bottleneck Is Agents Can't Coordinate

Economists calculate that double-digit growth requires an idea production function with returns greater than 1, but current data shows it well below 1—the real inflection point is not smarter models, but whether agents can coordinate like human teams.

AI economicsGDP growthAgent coordinationAnthropic valuationCompute regulationLabor markets

The video won't play here. Listen to the audio instead:

Two economists use economic theory and market data to dismantle the narrative that AI will deliver double-digit GDP growth, with high information density and almost no filler.

The argument · tap a timestamp to hear it

3:06

Benchmark hockey sticks don't match economic hockey sticks

Alex notes that meter reports show capability gains concentrated on well-defined tasks, while most economic work is not. Each model iteration should really ask: what previously impossible, economically valuable tasks does it actually unlock? A hockey stick curve on benchmarks doesn't mean the real world sees equivalent acceleration immediately—he thinks the real world is indeed accelerating, but the magnitude is exaggerated by benchmark numbers.

— Alex
6:13

Prediction markets suggest Anthropic's IPO valuation is too high

Tyler says over the past week he's become slightly more bearish on AI, partly because prediction markets on Anthropic's IPO keep fluctuating but never break 2 trillion. Looking at Anthropic's own revenue forecasts, while not disappointing, they're far from the hypergrowth narrative those valuations imply. He sums it up: institutions don't run as smoothly as people hoped or feared.

— Tyler
7:16

No plausible industry justifies these AI revenue projections

When Tyler pushes back on extreme OpenAI and Anthropic revenue forecasts in the Bay Area, he asks which industries will grow fast enough to validate those numbers. No one tries to answer. He concludes these extreme projections are false, though he doesn't think the whole industry is overblown—just that interest rates, stock prices and other market signals still look ‘quite normal,’ not the anomalies Bay Area people imagine.

— Tyler
12:24

Double-digit growth needs returns greater than 1, and data says no

Following papers by Ben Jones and Chad Jones, hitting ‘singularity-style’ double-digit growth needs one thing: the idea production function can't have diminishing returns—the parameter must exceed 1. Otherwise growth stays in the normal single-digit world. Alex says current data, including from meter reports, estimates this parameter well below 1: now you need to invest more resources to squeeze out the same intelligence gains, not less.

— Alex
16:33

Automation shrinks the economic contribution of what it automates

Alex uses non-homothetic preferences to explain a counterintuitive fact: when a technology gets widely automated and becomes cheap, its share of the total economy actually shrinks, so its contribution to GDP growth shrinks too. Value shifts to whatever bottleneck remains scarce. This means even with robots everywhere, the fastest-growing parts of the economy won't be the parts robots do—they'll be whatever stays scarce and un-automated.

— Alex
31:54

Single agents lose to human organizations because they can't coordinate

In a paper with DeepMind colleagues, Alex argues the real competition isn't AI agents versus individuals—it's AI versus human organizations. Organizations use comparative advantage to place each person in their strongest role, reaching collective intelligence far above average. Even if AI hits ‘minimal AGI’—average human performance on every task—the economic impact is limited. But testing specifically for coordination reveals a gap: one agent working alone beats two agents with communication between them; two humans, by contrast, far exceed one human doing two jobs. This means today's ‘agent swarms’ are just parallel processing. If the coordination ceiling breaks, firm size and hiring change radically.

— Alex
34:58

Rapid transitions risk trapping workers in scar-effect unemployment cycles

Alex's worst-case scenario: if coordination breakthroughs arrive, wages and employment undergo a ‘phase transition.’ Even if new sectors and opportunities appear, the worker reallocation that normally takes time may not happen fast enough, leaving people stuck unemployed and pushed out of the labor force—falling into vicious scar-effect loops. He stresses this is low-probability but high-cost if it happens.

— Alex
56:01

Jensen's troubles prove alignment can fail even inside institutions

On the Nvidia drama around Jensen Huang, Tyler half-jokes that ‘we already said alignment is impossible’—the company's behavior looks indeed misaligned, but checks and balances ultimately work out. The real danger isn't short-term but structural: if macro problems hit and Nvidia needs to serve as lender of last resort, a weakened or absent founder-CEO saps the firm's crisis response.

— Tyler

In their own words · checked verbatim

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.

Tyler6:13

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.

Tyler7:16

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, single-digit growth world.

Alex12:24

the value in the economy of what contributes to growth moves to the bottlenecks.

Alex16:33

you could actually show that one agent does better than two agents splitting the work, even if they can communicate.

Alex32:55

there is the potential for wages and employment to suddenly there's like a phase transition as far as what will happen to the economy.

Alex34:58

If you ask me the question, when will agents be 10% of GDP? I don't think it's soon.

Tyler37:06

Figures

Anthropic IPO prediction market valuation ceilingbelow 2 trillion6:13
Guyana annual growth rate40%13:27
Brief US growth after 1980s recessionabove 8%13:27
Tyler's forecast: AI agents' GDP share in 5 yearsabout 1.5%37:06
Elon Musk political spending in this election500 million42:21

Glossary

idea production function
An economic model describing the relationship between resource inputs and innovation or growth outputs.
non-homothetic preferences
A preference structure explaining why goods that become automated and cheap shrink their share of the economy.
Lucas critique
The warning that parameters estimated from historical data may not be ‘structural’ and can fail when policy changes.
mechanism design
Institutional arrangements that incentivize individuals to act in a principal's interest without requiring them to be naturally aligned.
mosaic intelligence
The collective intelligence an organization achieves when each person is placed in their most capable role, exceeding average individual intelligence.

How to listen

Who it's for

Investors and founders tracking AI valuation bubbles and macro growth narratives who want to assess AI's real GDP impact through economic frameworks rather than Silicon Valley rhetoric.

Skip

Alex's personal story about emigrating from Moldova and leaving medicine for economics (44:27-47:46) can be skipped.