weekend ai reads for 2026-03-20

In this episode, Harvard Business School professor Tsedal Neeley shares what she’s learned about successful AI implementation and organizational transformation, from the minimum technological capabilities needed to what it takes to overcome silos to how to transform workflows and processes to add real value.

AI tools are moving to consumption-based pricing, which means managers are going to have to think about how much money to invest in each individual. This is a massive paradigm shift. It’s like if you had to decide every month how good of a laptop each person on your team gets, and sometimes people run out of laptop halfway through the month.

AI isn’t replacing that work. Rather, it’s amplifying everything around it. The real shift happening is that designers are moving from being makers of outputs to directors of intent. From creators to curators. From hands-on executors to strategic decision-makers.

The AI wave isn’t caring about your org chart. It isn’t slowing down while you restructure. The window to get the right people doing the right work is open right now and it won’t stay that way.

Most excellence is merely just creating a plan and sticking to it.

[LinkedIn] has always been a disaster but I wonder what it will feel like when it’s a meta-disaster of bots writing and bots replying. In some ways it’s already the human equivalent of Moltbook, so maybe it’ll be better.

Clinical use of AI continues to grow in both prevalence and scope. Over 80 percent of physicians now use AI professionally, doubling since 2023. The average number of use cases per physician is 2.3 in 2026, up from 1.1 in 2023.

Courts will soon wrestle with questions such as how exactly AI affects what constitutes “readily ascertainable” information that therefore isn’t secret enough for protection.

When a buyer asks about something you listed, the technology scans your post and drafts a response using the details you already entered. The AI confirms availability, restates the price, and even includes the pickup location if you set one.

I’m software bonkers: I can’t stop thinking about software. And I can’t stop building software.

I asked my AI agent how it wants to remember things. It redesigned its own memory system, ran a self-eval, diagnosed its blindspots, and improved recall from 60% to 93% — for two dollars. The interesting part isn't the benchmark. It's what happens when you treat an AI as a participant in its own cognitive architecture.

LLMs optimize for plausibility over correctness. In this case, plausible is about 20,000 times slower than correct.

And while it is possible that AI itself could serve this function, AI still suffers from a trust deficit—most boards would still rather put their faith in advice from McKinsey or BCG than ChatGPT. (A more cynical take: CEOs still like to use consultants to justify their own decisions to boards, as well as to have someone else to blame if it all goes wrong.)

Adding more AI agents makes most systems worse. Three architecture patterns separate the $60M wins from the 40% that get canceled.

A three-hour workshop presented by Simon Willison at NICAR 2026.

Coding agents such as Claude Code and OpenAI Codex are mainly marketed at developers, but they’re actually applicable to a much wider array of problems, including data analysis, data cleaning, web scraping and other tasks commonly faced by data journalists.

They aren’t just faster versions of PostgreSQL or MySQL; they represent a fundamental rethink of how databases should work in an agent-driven world. Let me walk through the core principles driving this transformation.

Humans have and always will be important for watching ads.

The use of the word “song” for instrumental music — that is, music that is not sung and hence is not a song — is borderline illiterate. It illustrates more than anything how the entire system is designed for pop songs. For music of the western tradition, the word “composition” or “work” or “piece” is used except, of course, if the composition is actually a song.

Sometime in the last twenty years, our possessions came alive.

Nothing you own is finished. Everything exists in a state of permanent incompletion, permanently needing.

I hope this email isn’t interrupting anything urgent.

Ostensible meaning: I am respectful of your time.
Actual meaning: I have no idea how email works.