ENCORE AI Series: Anthropic's Peter McCory

The episode begins with Moody's Analytics Chief Economist Mark Zandi and co-hosts Marissa Di Natale and Chris Reades discussing recent economic data and Fed policy. The Federal Reserve held rates steady but faced internal debate about future policy communication, with Jerome Powell announcing he will remain on the board of governors indefinitely due to concerns about Fed independence. Recent economic data shows weak GDP growth at 2.6% (down from 3.1% estimates), elevated inflation with core PCE at 3.2% year-over-year against a 2% target, and a historically low savings rate of 3.6%, all amid rising oil prices and geopolitical tensions.

Peter McCory, head of economics at Anthropic, joins to discuss Claude's economic impact. McCory shares his unconventional career path from philosophy and English major to economist, working at the St. Louis Fed, completing a PhD from Berkeley focused on fiscal policy, and subsequently working at JP Morgan and LinkedIn before joining Anthropic to develop the Anthropic Economic Index.

The hosts and McCory discuss practical applications of Claude in economic research. Mark Zandi describes using Claude to evaluate housing legislation, asking it to write in his voice and provide sensitivity analyses—saving roughly 30 hours of work. Chris Reades highlights Claude's role in learning Python and iterating on economic models. Marissa Di Natale describes using Claude to improve labor market indices by testing different permutations of calculations. McCory emphasizes two key value dimensions: making existing work more efficient and broadening the scope of what economists can accomplish, with data showing longer-term Claude users achieve greater value through more collaborative interaction patterns.

On AI's macroeconomic impact, McCory presents Anthropic's analysis suggesting AI could contribute 1.8 percentage points annually to labor productivity growth over the next decade, compared to consensus economist estimates averaging 0.8 percentage points. This is based on observed task-level time savings of 80-90% across Claude users, scaled through the economy's occupational structure using Halton's theorem. However, McCory qualifies this number, noting that accounting for task failures and bottlenecks where human expertise remains essential could reduce the estimate closer to 1 percentage point labor productivity lift. He emphasizes that current analysis reflects today's models and usage patterns, and that faster model improvements could alter these projections significantly.

McCory argues AI diffusion may occur 5-10 times faster than past transformative technologies because it requires no specialized infrastructure and can be accessed immediately. He notes that economist forecasts for AI's productivity contribution range widely (roughly 0.5-1.8 percentage points for TFP/labor productivity), using historical analogs like the internet era as reference points. He points out that long-run U.S. GDP per capita growth has averaged 2% annually over 150 years despite massive structural transformation, suggesting risks to very optimistic scenarios.

Regarding labor market implications, McCory presents data showing no material unemployment impact yet for workers in occupations with high AI exposure, but notes hiring rates for younger workers in AI-exposed roles have weakened in the past year. Survey data from 81,000 Claude users reveals that workers in occupations where Claude is used for automation express greater job displacement concerns, particularly younger workers. However, McCory argues the distinction between AI as a substitution technology versus complementary technology remains unclear—current evidence suggests AI may be skill-biased, amplifying productivity gains for those with requisite expertise while potentially displacing workers in routine tasks like data entry and customer support.

McCory emphasizes that business cycle dynamics matter significantly. He references research showing recessions accelerate technology adoption as firms restructure operations, potentially amplifying downturns if AI adoption accelerates during economic weakness. Conversely, with appropriate macroeconomic policy maintaining full employment and clear price signals, workers can reallocate to areas of demand despite sectoral disruptions.

On underemphasized benefits, McCory highlights that AI's potential extends beyond GDP growth to accelerating scientific innovation and improvements in health and human flourishing—potentially solving problems that might otherwise take decades to address.

Zandi raises the philosophical Ship of Theseus question about repeatedly using Claude to generate content in his voice, questioning when the output becomes Claude's work rather than his own. McCory responds that ongoing human oversight and tacit knowledge remain essential—the complexity of tasks determines success rates, and enterprise adoption data shows that complex automation requires disproportionately more contextual information than simple tasks, reflecting the continuing complementarity between human expertise and AI capabilities.