Issue 1 · November 10, 2026
Good alone, not yet good together
Four things worth reading on what the newest models can do, why the labor market hasn’t moved much, and what that means for the people we train.
From the Director
When Dean Chatterjee introduced Terry’s Strategic Plan for AI Integration in August, one of the commitments was a regular place to share what is happening in AI and how our colleagues are putting it to work. This is that place.
Each issue will have three parts. First, a few things worth reading, with summaries short enough to help you decide where to spend your time. Second, when we have one, a story about how someone at Terry is using AI in research or in the classroom. Third, what is coming up from the Institute and the Terry Analytics Lab.
This issue’s reading has a common thread. The models are now good enough that the interesting questions are about organizations and jobs, not benchmarks. Two economists come at that from different directions, and both land on a problem we own: how to train people for jobs whose entry-level tasks a machine can do.
Worth Reading
Alex Imas, Andrew Koh, and Shane Legg · DeepMind Institute · Oct. 29
The Economics of Intelligence: How and when does AGI lead to economic transformation?
AI capabilities are improving fast, but employment and productivity have barely moved. The authors argue we are measuring the wrong thing. Benchmarks score a single agent working alone; economic outcomes come from organizations. Gains in competence change the economy gradually. Gains in coordination, which today’s agents still lack, would change it abruptly.
Why it matters. Their earliest warning signs are wages and job roles, not GDP. Flat employment numbers are weak evidence that nothing is happening.
Jason Furman · Econvitals · Nov. 1
AI Could Do My First Job. Not My Last.
Furman defends his remark that, running the Council of Economic Advisers today, he would reluctantly choose current AI over 20 junior staff. Exposure depends on what a job consists of, not on seniority or talent. AI could have done most of his work as a junior economist and little of his work as Chair.
Why it matters. He names the pipeline problem directly: if juniors produce little, who pays to train the next generation of seniors? That is our business.
Ethan Mollick · One Useful Thing · Nov. 9
Fable 5.2 and Jagged AGI
Mollick gave the newest Claude model hard tasks across very different fields, including an original entrepreneurship paper he rates ahead of a typical second-year PhD student. The failures were not hallucinations. They were lapses of judgment and taste: a 182-page paper, because no critique was ever answered by cutting anything.
Why it matters. Knowing when your own work isn’t good enough is the skill the model lacks, and the one we should be teaching.
Addy Osmani · X · Oct. 29
You’re probably overprompting Claude
Newer models need far less instruction than most of us still give them. Osmani’s advice: state the goal, give the model a way to check its own work, and delete the instructions you wrote to get around the limits of older models.
Why it matters. A practical one. If your prompts are a year old, they are probably holding you back.
From the Institute
AI in Staff Workflows
Sadie Brown, Finance & Administration, Dean’s Office · Noon, Orkin D207 · Register
Obsidian for AI Workflows
Laura Zimmerman, Associate Professor of Economics · Noon, Orkin D207 · Register
Have something we should read, or a story about your own use of AI? Reply to the listserv message and let me know.
Chris