READING LIST — OPTIONAL RESOURCES

Ethan Mollick reading list

A curated list of Mollick's One Useful Thing essays, organized by theme. Browse by topic based on where you are in the course — you don't need to read all of these.

Ethan Mollick is Associate Professor at the Wharton School of the University of Pennsylvania. His Substack, One Useful Thing, is the most consistently practical writing on working with AI. Several ideas from this course — models / apps / harnesses, cyborg vs. centaur, "you aren't prompting, you are managing" — come directly from his work.

Getting started / foundations

An Opinionated Guide to Using AI Right Now

October 19, 2025

Which tools to use, which to skip, and why the gap between marketing and actual capability matters. Practical selection criteria for journalists evaluating new tools.

Getting started with AI: Good enough prompting

November 24, 2024

Argues that context matters more than prompt perfection; recommends 10 hours of hands-on practice over studying theory. Directly relevant to understanding why context files (CLAUDE.md, GEMINI.md, AGENTS.md) work.

A guide to prompting AI (for what it is worth)

April 16, 2023

The classic Mollick prompting guide. No magic prompts exist; what works is giving AI context (persona, audience, constraints, examples) and iterating. This is the conceptual foundation for Module 2.

15 Times to use AI, and 5 Not to

December 2024

AI is most useful for tasks you could do but shouldn't waste time on. Identifies where AI genuinely helps vs. where it undermines judgment or quality. Read before designing any automation.

Working methods: cyborg vs. centaur

Captain's log: the irreducible weirdness of prompting AIs

March 4, 2024

Different prompt framings (Star Trek, political thriller, etc.) dramatically change AI accuracy. Context isn't just about facts — it's about establishing a frame for how AI should think. Directly relevant to how you write CLAUDE.md files.

Innovation through prompting

April 22, 2024

Argues that prompts function as "programs in prose" that non-technical experts can author. The conceptual basis for custom skills: you don't need to write code to encode domain expertise.

Management as AI superpower

January 27, 2026

Managing AI agents requires the same skills as managing people: clear instructions, feedback, and evaluation criteria. As development shifts from writing code to directing AI, management skills become the bottleneck.

Giving your AI a Job Interview

November 12, 2025

How to test and customize AI models to find what works for your specific tasks. Useful for evaluating whether a tool fits a particular reporting workflow before committing to it.

Agents and automation

Real AI Agents and Real Work

September 29, 2025

AI agents that can plan and use tools to accomplish multi-step tasks are now practical, not theoretical. Small accuracy improvements lead to large gains in what agents can complete. Read before Module 4.

On Working with Wizards

September 11, 2025

Working with AI systems that produce impressive results but operate in opaque ways. Covers trust, verification, and workflow design — critical for journalism, where you can't just accept output.

Three Years from GPT-3 to Gemini 3

November 18, 2025

Chronicles how fast model capabilities have changed; demonstrates Gemini 3 building complete interactive applications. Useful context for understanding what agent-based workflows can now do.

AI's capabilities and limits

On Jagged AGI: o3, Gemini 2.5, and everything after

April 2025

Introduces "Jagged AGI": AI with superhuman ability at narrow tasks but obvious gaps elsewhere. Explains why AI fails in unpredictable ways — and why retrieval systems (RAG) matter: they fill the gaps the model can't fill from memory.

Mass Intelligence

August 28, 2025

We're entering an era where powerful AI is as accessible as Google Search. Every institution built for scarce intelligence — schools, hospitals, newsrooms — now has to operate with abundant AI. Raises the stakes for figuring out what human editorial judgment is actually for.

GPT-5: It Just Does Stuff

August 7, 2025

Newer models automatically select computation depth and handle complex tasks with minimal instruction. Relevant for understanding why describing a workflow in plain English can produce a working pipeline.

Education and learning

Post-apocalyptic education

August 30, 2024

Addresses what education looks like when AI can complete most traditional assignments. The parallel to journalism is direct: if AI can generate copy, what is a journalist's editorial judgment actually for?

Book

Co-Intelligence: Living and Working with AI

Penguin Random House, 2024

NYT bestseller; named best book of 2024 by The Economist and Financial Times. Develops the cyborg/centaur framework for human-AI collaboration. The conceptual foundation for everything in this course.

Already assigned in the course

These Mollick pieces appear as required readings in specific modules:

A guide to which AI to use in the agentic era

INTRODUCTION MODULE + MODULE 01

Models / apps / harnesses framework — the foundation for how the course talks about AI tools.

Claude Code and What Comes Next

INTRODUCTION MODULE

What the harness layer looks like in practice.

NEXT: Head back to Module 1 or browse the full syllabus.