🔮 Exponential View #562: Agents & the tedium frontier; AI in the statistics; robot insurance; Claude at war, hacking pigeons, AI dignosis++
Azeem is the GOAT of AI analysis. – Max P., a paying member
Hi all,
Welcome to the Sunday edition. Today’s email stays unusually tightly focused for us: wages, robots, and token costs – and how, together, they’re beginning to rewire the basic economics of work.
Enjoy!
The tedium frontier has moved
I’ve been living with my always-on AI agent, R Mini Arnold, for a couple of weeks now. What it’s already made clear is that once you drop the transaction cost of delegation by an order of magnitude, a personalized agent becomes infrastructure for knowledge work – in my case, it’s cleared a backlog of tasks that would otherwise never get done. I write about this in my weekend essay and alongside that reading, my conversation with Rohit Krishnan digs into how we both use agents to write and think:
The productivity staircase
After years of the “Solow Paradox 2.0”, seeing AI everywhere except in the productivity statistics, early 2026 is an inflection point. Erik Brynjolfsson argues that we’re seeing “a US productivity increase of roughly 2.7%.” Revised BLS data show “robust GDP growth alongside lower labor input” signals that the economy is transitioning out of the “investment valley” and into the “harvest phase”. Alex Imas points out that recent micro-studies consistently demonstrate significant performance boosts from generative AI, even if not yet widely reflected across the economy.
Most firms remain in shallow adoption. A new study by Nicholas Bloom and colleagues shows that, despite 70% of firms claiming to use AI, senior executives spend on average just 1.5 hours a week with these tools. Around 20% of firms report productivity gains, but the remaining 80% do not, so the aggregate impact on productivity over the past three years is essentially flat.
My view is that these micro gains are now turning into broader gains. Our own data shows that American public firms more frequently cite quantitative success measures when discussing their genAI projects. But I also think these will staccato across the economy, unevenly, dependent on firm leadership, access to capital, workforce capability and other factors. From a distance, the curve will be smooth; up close, it’ll be a juddery staircase.