Advanced AI Agent Design

Target audience

This course is designed for technical professionals and AI Champions who have already taken our “Design of AI Agents” course, or gained corresponding knowledge in other ways, and want to learn more advanced topics (see topics below). Participants should have some hands-on experience with building AI agents.

Ideal for: AI Architects, Senior Developers, Technical Leads, Solution Architects, and AI Specialists responsible for building production AI agent systems.

What you will learn

The exact agenda is flexible, but we typically cover topics like:

• Advanced context management - how to make sure the agent gets the context it needs, neither too much nor too little, and how to dynamically optimize the context using techniques like subagent delegation.
• Multi-agent architecture - when and how to split a workflow into multiple agents, and how to set up effective orchestration patterns and collaboration between agents
• Information architecture - how to work effectively with large documents, databases, and external sources of information, using techniques such as RAG (retrieval augmented generation) and document indexing.
• State management - how to handle complex multi-state workflows
• Memory management - how to work with memory to let the agent build up a persistent context over time
• Safety - how to work with transparency and guardrails to ensure the agent doesn’t leak information or cause damage.
• Custom tools and integration - how to integrate with enterprise systems, and how to use http and protocols like MCP (Model Context Protocol) to dynamically extend the agent’s capabilities.
• API interaction - how to let an agent interact with external APIs, and how to your agent be exposed as an API to your services
• Production readiness including security, monitoring, and system integration
• Testing and evaluation - how to systematically evaluate how the agent is performing
• Code and app generation - how to automate predictable parts of the agent’s workflow using scripts, and how to dynamically create apps and dashboards to improve human-agent collaboration.

Course format

The course is based mostly on live demos, examples, discussions, and small exercises. No death by powerpoint! All content is tied to real-life experience rather than theory.

We will primarily use the Abundly agent platform for exploring and demonstrating advanced concepts. But we will also discuss other platforms, and the topics covered are relevant regardless of which agent platform or ecosystem you are working with.

Prerequisites

• Completion of "Design of AI Agents" course or similar knowledge from another course, or by doing a lot of experimentation on your own.
• Some programming experience with APIs, system integration, and basic software architecture. No need to be an expert.
• Active use of AI agents in a professional context.

Course instructor

Henrik Kniberg

Format: 1 day, onsite in Stockholm (Swedish), or remote (English) Price: Early Bird [price] SEK | Standard [price] SEK (excl. VAT)

What you get

• Access to all course materials, templates, sample agents, and frameworks
• Extended access to Abundly's agent platform (app.abundly.ai) with enterprise features
• Course certificate confirming your advanced AI agent architecture competencies
• For onsite course: Lunch and refreshments including morning sandwich and fruit

Frequently asked questions

Q: How does this differ from the basic agent design course?A: This course assumes you already know how to build basic agents, so we jump in to the deep end of the pool.

Q: How does this differ from the agent hackathon

A: The hackathon is only half a day, and is entirely centered around you building your agents, with our help. The advanced agent design course is a full day and more focused on advanced topics and smaller demos and exercise around those. These are complementary workshops, you will benefit from attending both, in any order.

Q: Do I need programming experience? A: At least some experience, yes. You should be comfortable with APIs, system integration concepts, and basic software architecture principles. But you don’t need to be an expert-level programmer.

Q: Will this work with non-Abundly platforms? A: The principles, patterns, and techniques discussed in the course apply regardless of platform. However the available feature set will of course vary depending on platform.