I spent 9 months building, testing, breaking, rebuilding, and shipping AI agents.

And I've finally got something to show for it (note: you'll love it)

But first, some context...

Most engineers can build an AI agent demo.

Very few can build one that survives production.

So I collaborated with Towards AI to close that gap.

We built two real production systems first.

Used them.

Stress-tested them.

Deployed them.

Then we turned the entire engineering stack into a course:

𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴.

The sole focus is to help you go from experimenting with agents to engineering production-grade systems.

You’ll learn how to design systems that:

• Stay reliable under real usage

• Handle state and long-running workflows

• Include evaluation from day zero

• Are observable and debuggable

• Deploy cleanly to the cloud

And by the end, you’ll have built:

• A fully deployed Research Agent

• A production Writing Workflow

• Custom evaluation systems with LLM judges

• Observability and tracing for non-deterministic behavior

• Docker + Cloud Run deployment

• Authentication, state management, CI/CD

More importantly, you’ll master the decision framework behind it all:

When to use a workflow vs an agent

When to introduce autonomy

When to keep humans in the loop

How to design systems that evolve without collapsing

This is system design for agentic AI.

If you want portfolio-ready proof that you can architect and ship AI systems in production, this is what we built it for.

Enroll here: academy.towardsai.net/c…

P.S. We opened 100 early-bird seats at $449. More than half are already taken.