Tyler Cowen: AGI is already here by the original definition, but human bottlenecks mean slow economic takeoff

Summary

Tyler Cowen argues AGI has already arrived under the original definition — a system that passes the Turing test and matches expert human performance on intellectual tasks — and that debating the label is less important than acknowledging the capability is here. He uses o3 and o3 Pro daily, describing them as superior to a specialist human guide for research and travel planning. The more consequential debate, in his view, is not whether AGI exists but how fast it translates into economic output.

Human Bottlenecks, Not AI Limits, Drive Slow Takeoff

Cowen's core thesis is that human institutional friction — not AI capability — will constrain the pace of economic transformation. He draws a direct parallel to simpler behavioral improvements like workplace culture that economists have identified for decades but that organizations still fail to implement. His speculative estimate for AI-driven GDP uplift is half a percentage point per quarter, which he concedes is enormous in aggregate but unlikely to show up dramatically in near-term market data.

He interprets current market pricing as evidence for this view. Nvidia's $4 trillion market cap reflects AI enthusiasm, but the fact that everything else trades at roughly normal multiples suggests investors are not yet pricing in a rapid economic transformation — consistent with a slow-takeoff scenario. He reads Satya Nadella's recent posture — layoffs, operational pragmatism — as aligned with the same thesis: transformative over time, not next quarter.

Physical-world constraints reinforce the argument. Most jobs intersect with the physical world, and robotics remain far off, which means the largest categories of labor displacement are structurally delayed regardless of software capability.

Developer Productivity Study Dismissed

Cowen flatly rejects a widely circulated study claiming AI coding tools produced a 19-20% productivity decrease among software engineers working on open-source projects. His objections are methodological — only 16 data points, and only one participant had prior AI experience, with that individual actually recording a productivity gain. He argues real-world marketplace evidence overwhelmingly supports productivity improvement and that the paper should be discarded. He personally reports AI cutting podcast preparation time in half.

AI Talent Wars and Compensation

On the spike in AI researcher compensation, Cowen expects normalization by 2030 rather than a continued escalation toward nine-figure signing bonuses. He compares superteam formation in AI labs to NBA roster construction — the Miami Heat, Chicago Bulls, and late-era Lakers all demonstrated that assembling elite talent does not automatically produce results, and that timing, chemistry, and trajectory matter as much as individual quality.

Windsurf Acquisition and Covenant Risk

The Google-Windsurf deal — where Google acquired roughly 50 engineers and paid out the CEO and preferred shareholders while approximately 400 employees were left in a stranded entity with roughly $150 million on the balance sheet, later absorbed by Cognition — illustrates a structural gap in startup financing contracts. Cowen frames it as a Kaldor-Hicks rather than Pareto improvement and warns it raises the cost of capital for the broader sector. The fix, in his view, is market-driven covenant development by investors, modeled on bond indenture protections that prevent asset-stripping. He explicitly opposes FTC involvement, calling both the previous and current leadership economically illiterate.

Stablecoins and Dollar Access

Cowen is constructive on stablecoins as a workaround for international demand for dollar-denominated systems, demand he says the US government itself partially suppresses through know-your-customer reporting requirements that make American account holders toxic to foreign banks. He supports the GENIUS Act, describing it as imperfect but sufficient to create a legal framework for experimentation. Long-run stablecoin dominance is uncertain given regulatory and competitive dynamics, but near-term momentum is real.

Public Markets and Corporate Structure

Cowen argues public markets are structurally disadvantaged relative to private capital for long-duration, high-uncertainty technology bets — too much disclosure, too much bureaucracy, shareholders prone to short-termism. He favors pure-play public listings over conglomerate structures like Elon Musk's emerging keiretsu across Tesla, SpaceX, xAI and Neuralink, on the grounds that separate tickers generate cleaner price signals. On Delaware, following Andreessen Horowitz's move to Nevada, his position is unambiguous: exits should continue until the state feels enough fiscal pain to reform.