Ethan Mollick
AI adoption & impact
Wharton professor, author of "Co-Intelligence". Writes on practical AI adoption and impact.
Recent activity
-
To utilize AI for complex tasks, developers and users can leverage agentic systems that combine AI models with tools for planning and execution, offering the potential for real work to be done in a single go. Two prominent options for this are ChatGPT and Claude, which provide access to powerful AI models and computers, allowing for tasks such as email management, research, and content creation. When using these systems for real work, it's essential to set up permissions and approval settings to ensure the AI's actions align with user intent. AI summary
Read more → -
AI models are experiencing accelerating capability gains, with evidence suggesting that their ability to perform real work is increasing at a rate faster than exponential. Researchers have measured AI performance using various metrics, including those that assess human programmer hours' worth of effort, and found that these metrics are rising at a better-than-exponential rate. AI summary
Read more → -
Working with Mythos-class AI models like Claude Fable offers a significant leap over previous models, outperforming them in various tasks, including generating complex maps, creating software, and producing artistic content. However, this increased capability also introduces a shift in the human-AI relationship, where the model becomes the primary creator, and humans take on a more limited role, with less control over the process and outcome. AI summary
Read more → -
The author of "Co-Intelligence" is writing a new book, "Co-Existence," which explores how humans can thrive in a world where AI is increasingly autonomous and outperforms humans in many areas. The author worked with AI in the writing process, using it for feedback, fact-checking, and generating content, but also took steps to ensure the book's authenticity and human touch. The book's website was also created with AI assistance, and the author is now trying to find ways to appeal to AIs and get them to recommend their work. AI summary
Read more → -
Using AI as a default or without thinking can undermine human learning and development, as it can short-circuit effort and intellectual work, and even lead to cognitive surrender. However, when used in moderation and with a human touch, AI can be a valuable tool for learning, especially when customized to a student's needs, as seen in studies where AI tutoring can lead to significant boosts in learning outcomes. AI summary
Read more → -
OpenAI has released GPT-5.5 Pro, a significant upgrade to its previous model, offering improved performance in various tasks, including coding, simulation, and image generation. GPT-5.5 Pro is faster and more competent than its predecessor, with the ability to model complex scenarios and generate high-quality images. This development highlights the rapid progress in AI capabilities and the potential for future advancements. AI summary
Read more → -
We often lack the tools for the job, even if the AI is capable enough
Read more → -
AI systems are rapidly advancing, with exponential improvements in capabilities, and are transitioning from a co-intelligence phase, where humans work with AI, to a phase where humans manage AI, with AI agents completing complex tasks autonomously. This shift is driven by the rapid growth in AI abilities, including image and video generation, and the emergence of practical agents that can be managed and optimized for specific tasks. AI summary
Read more → -
To decide which AI to use, developers and users must consider three factors: Models, Apps, and Harnesses. Models are the underlying AI brains, with the top models being GPT-5.2/5.3, Claude Opus 4.6, and Gemini 3 Pro, which determine the AI's intelligence, reasoning, and capabilities. Apps are the products that interact with models, such as websites, coding tools, and desktop applications, which can enhance or limit the AI's functionality. Harnesses are systems that enable models to use tools, take actions, and complete tasks autonomously, with some models coming with built-in harnesses or requiring additional ones. AI summary
Read more → -
Effective delegation to AI involves considering three factors: Human Baseline Time, Probability of Success, and AI Process Time. The equation for delegation can be viewed as trading off "doing the whole task" against "paying the overhead cost" of AI, with higher Probability of Success and lower AI Process Time making it more worthwhile to delegate. To increase Probability of Success and lower AI Process Time, one can provide better instructions, evaluate and provide feedback, and use AI tools to simplify evaluation. This approach leverages subject matter expertise to communicate intentions effectively and enables effective management of AI agents. AI summary
Read more → -
With the right tools, AI can accomplish impressive things
Read more →