This talk explores what it means to work with AI from inside a live Pharo system rather than treating AI as an external chatbot. I will demonstrate experiments where AI models are invoked directly from the Playground, including a concrete implementation of conversation history (AIAHistory) and experiments across multiple models and providers.
Rather than focusing on code generation, the emphasis is on workflows: how different conversation structures affect results, how styles and constraints can be imposed on generated comments, and how AI systems can be inspected for systematic errors and limitations. Examples include documentation support, UML generation (via PlantUML), and experiments in automated paper review. The talk is experimental in nature and rooted in Smalltalk’s strengths: reflection, live objects, and tools that are part of the system rather than bolted on. The goal is not to present a finished framework, but to share concrete insights, failures, and possibilities for AI as a native Pharo tool. Kasper Østerbye ( https://github.com/kasperosterbye ) is a retired computer science researcher with a long background in programming languages and live systems, including decades of work with Smalltalk and Pharo. He now explores how AI models can be integrated as native tools inside a running Pharo system, focusing on conversation structure, systematic failures, and tool design rather than automation alone. This will be an online meeting. If you'd like to join us, please sign up in advance on the meeting's Meetup page ( https://www.meetup.com/ukstug/events/311265921/ ) to receive the meeting details.
