Last weekend at Tulane, Harry Wang (Delaware) pointed out that a year ago, 25% of Y Combinator startups had AI-generated code. Today it's 100%. The term "vibe coding" didn't even exist 12 months ago. Lovable, built around it, hit a $6.6B valuation in a year.

Jon Krohn gave a thought-provoking keynote at Tulane this weekend. In Jeanne Calment's lifetime, she saw the invention of the lightbulb, TV, penicillin, space flight, and the internet. Literacy went from rare to nearly universal, and lifespans doubled. What will happen in our own lifetime?

AI has an amazing potential to increase productivity. Will the gains diffuse to developing countries the way mobile phones did? I think it will. The computation happens off site in data centers thousands of miles away. The infrastructure doesn't need to be local!

Another paper finds internet access boosts grain production A new paper in the Review of Development Economics asks whether having access to the internet and smartphones makes farmers more productive.

Another paper finds that internet technology increases grain production. Amazing how something so new can affect something so ancient. buff.ly/F1XNJKp

When Honesty About Seeds Discourages Farmers Miehe et al.

When told that high-yield seeds need fertilizer, precise timing, and careful management, Ugandan farmers became less likely to plant them. Some researchers call that a failure. I think it sounds like a rational choice. buff.ly/JqhZOGM

I had the honor of hearing Sir Tim Berners-Lee speak at the Tulane Bookfest yesterday. He said that casual conversations in the CERN staff lounge helped inspire the World Wide Web. Now he worries the Web has isolated us instead of connecting us. Such a remarkable and humble man.

When Cooperatives Work Well A new paper by Oduniyi, Chike, and Antwi (2026) in Scientific African asks whether membership in agricultural cooperatives improves the lives of smallholder maize farmers in Mpumalanga, South Africa.

Enjoyed digging into a new paper by Oduniyi, Chike, and Antwi on agricultural cooperatives in South Africa. The cooperatives clearly help. But the data also tells a Coasian story about economic institutions. Blog post here:
buff.ly/lztbIsY

Mobile Phone Sharing in Togo I’ve been thinking a lot lately about how technology is shared, and how this sharing helps with economic development.

Cell phone sharing is common in the developing world. A paper out of Berkeley documents this pattern for Togo, and shows how sharing offers solutions and challenges to distributing development aid.
buff.ly/cnUl9yN

Which digital tools actually pay off for small businesses? Not all of them. Digital tools make firms better at acting on opportunities and reorganizing around them. They don't help firms identify new opportunities. That apparently requires actually talking to customers. buff.ly/sIxqCCx

Saudi Women Show Us What Happens When Government Gets Out of the Way A new paper in Information Technology for Development by Alotaibi, Dasuki, and Zamani (2025) documents what happened when Saudi Arabia relaxed its male guardianship system and launched mobile…

Saudi Arabia lifted its guardianship laws and women immediately opened bank accounts, started businesses, and got PhDs. They just needed their government to get out of their way. An app helped smooth the process.
buff.ly/92kn8Zd

Switching to three in-class exams means I need to skip at least three days of class-time lecture. Maybe more. It hurts my heart.

Mobile Money as Anti-Corruption Tool Laura Barasa’s 2025 study in Information Technology for Development estimates the effect of mobile money adoption on bribe payments in Kenya.

Kenya's M-PESA accidentally became more effective at fighting corruption than the official Anti-Corruption Commission. New research shows that mobile money reduced bribe payments and the time spent on regulations. Full analysis on my Substack:
buff.ly/bD5MasV

Professors, how much of your daily routine have you been able to fully automate with AI/coding. Me, almost none. I'm looking for ideas.

The (Indigenous) Knowledge Problem in Development Economics There’s a curious overlap across ideological lines that few people seem to have noticed.

What do F.A. Hayek and postcolonial development scholars have in common?
A lot more than I thought.

Prediction: Thanks to AI, academics will publish an exponentially greater number of research papers than ever before. Universities will soon double down on quality over quantity. AI automates writing routine correlational statistical papers. Their value will plummet.

Collapse Your Instruments when using GMM This one’s a bit inside baseball, so apologies in advance to readers who don’t regularly estimate dynamic panel models.

New on Substack: "Collapse Your Instruments when using GMM". If you use Arellano-Bond GMM estimators, you're probably using too many instruments and getting biased estimates. Here's a short post on what to do about it. [https://buff.ly/qXfXdSH]

Default and collapsed instruments in dynamic panel GMM: a Monte Carlo comparison

Happy to announce that my paper on dynamic panel GMM estimation was just published in Economics Bulletin. I show that using a collapsed instrument set outperforms the default settings across virtually all panel configurations of panel GMM. buff.ly/sFNyUwR

I miss using em-dashes. Thanks, AI.

Papers Worth Reading: Chen (2026) and the Limits of Economic Planning In my previous post, I summarized Yutong Chen’s 2026 paper on how India’s demonetization affected firms differently across sectors (click here). Service firms in digitally-prepared districts gained;…

The Indian government had a plan. The economy had other ideas. Check out my new blog post on Chen (2026), Hayek, and why large-scale interventions produce surprises.
johnlevendis.substack.com/p/papers-wor...

Papers Worth Reading: Chen (2026) on Digitalization’s Uneven Effects in India

India's digital currency push its service firms and hurt manufacturers, in the same cities, and at the same time. Yutong Chen's new paper shows why averages lie. johnlevendis.substack.com/p/papers-wor...

The hardest part of teaching in 2025 isn't the material. It's figuring out how to teach when every assignment can be outsourced to a chatbot.

Spring semester is coming and I'm prepping my new Machine Learning class. Teaching AI, testing with pen and paper. LoL Anyone else switching back to in-person exams?

Why Voting Won't Save AI Ethics Either In my previous post, I explained how two ethical principles most people accept (individual autonomy and unanimous agreement) can be logically incompatible.

Granted, we can't all agree on specific AI ethics principles. But why can't we just vote to get something the majority of us can agree on? Here's why...
open.substack.com/pub/johnleve...

Why We Can't Just Program Ethics Into AI We agree: we should hard-code ethics into the AIs.

In 1970, Amartya Sen proved something that makes AI alignment much harder than most people realize. Two principles everyone agrees on (respect autonomy + respect unanimous preferences) can be mathematically incompatible. No algorithm can fix that.
johnlevendis.substack.com/p/why-we-can...

Just presented my paper on "IT and Stock Market Efficiency" at the GlobDev workshop at #ICIS2025. Great ideas on IT as a tool for political resistance and resilience. Thanks to @SajdaQ and Pitso Tsibolane for organizing. 10/10 would go again.