A hands-on workshop led by George Mandis
Every week there's a new AI agent framework, a new SDK, a new tool promising to revolutionize how your team works. But underneath all of them are the same small set of building blocks: a model, a system prompt, tool definitions, and structured outputs.
Most engineering leaders are making decisions about these systems without a clear mental model for how they actually work. The result is decision-making driven by vendor marketing, demo theater, or vibes.
This workshop fixes that. In 80 minutes, you'll build AI agents from scratch — starting with a bare model and progressively adding capabilities — and develop lasting intuition for how the pieces compose.
Using the open-source Composable Agent Playground — a tool I built specifically for this workshop — you'll work through a series of guided exercises:
Each exercise is designed to surface a specific "aha" moment about how these systems actually work. You'll toggle tools on and off, swap between models, edit prompts, and constrain outputs — observing how each change affects the agent's behavior in real time.
What a model actually does (and doesn't do). Why model selection matters less than you think. How to swap providers — OpenAI, Anthropic, Ollama — and see the same tools produce different results.
How a few sentences of instruction shape an agent's entire personality and workflow. You'll make small changes and watch dramatically different outcomes emerge. This is where most of the real "engineering" in prompt engineering lives.
This is the key mechanism that turns a chatbot into an agent. You'll add and remove tools — search, data lookups, UI interactions — and see the model decide when and how to use them. You'll inspect the JSON Schema definitions the model actually sees, and understand why tool design is an API design problem.
How schemas constrain responses into predictable, machine-readable formats that your applications can reliably consume. The difference between hoping for JSON and guaranteeing it. This is what makes agents composable — not just conversational.
Engineering leaders, tech leads, staff+ engineers, directors, and anyone making decisions about AI adoption for their team or organization. You don't need an AI or machine learning background — just curiosity and a laptop.
Whether you're evaluating an AI coding assistant for your team, building internal tools with LLM capabilities, or just trying to have an informed opinion in your next architecture review — understanding the composable building blocks of agentic systems is immediately practical and increasingly essential.
80 minutes. Interactive and hands-on. Participants work in small groups through guided exercises.
Available as:
I'm George Mandis — a software consultant, fractional CTO, and creative technologist based in Brooklyn with over 20 years of experience building software and leading engineering teams. I've worked with organizations including Axios, the NBA, and the World Wildlife Fund.
I run Less Software, an independent practice focused on simpler systems, engineering effectiveness, and AI enablement. I built the Composable Agent Playground that powers this workshop because I care more about helping people develop lasting intuition than about any particular framework or vendor.
I've been leading workshops at conferences for years — on topics from WebMIDI to Arduino to progressive web apps. This one is different because the need is so urgent: engineering leaders are making consequential decisions about AI adoption every day, often without hands-on understanding of how these systems are actually built.
Interested in bringing this workshop to your team or event? Email me, connect on LinkedIn, or schedule a call to discuss.