Building Your Own Agent Experience with Pi

A practical two-hour exploration of how developers can move beyond the fixed behavior and interfaces of conventional coding assistants and create agent experiences tailored to their own tools, workflows, and applications.

Using Pi as the implementation context, we will examine how a small agent core can be extended with new capabilities, interactive interfaces, custom workflows, and programmatic behavior without requiring a large agent framework.

We will also explore what happens when that agent is no longer tied to one terminal session: when work can persist, resume after interruption, be observed from different interfaces, or become part of a shared application.

The emphasis is on the underlying ideas that make an agent adaptable: extensibility, composition, embedding, state, and durability.

Audience: developers at any level who use AI coding tools and want to understand how those tools can become programmable building blocks rather than fixed products.

Outcomes:

Understand how an agent can expose its behavior as an extensible software surface

Explore custom tools, interactions, and workflows

See how agent functionality can be embedded into another application

Understand programmatic composition of agent tools and capabilities

Explore persistent conversations and resumable agent work

Understand the possibilities created by shared and multi-interface agent sessions

Identify practical situations where a lightweight programmable agent can replace a more rigid workflow

Format: two hours with a fast conceptual walkthrough, compact demonstrations, an interactive end-to-end example, discussion, and questions and answers.