# Agent learning
In November 2025, over 1.5 million people signed up for a free course on building AI agents. This is not a typo, nor is it a vanity MOOC indicator for sign-ups that never convert: the same cohort reported over 11,000 culmination submissions. In terms of number of participants, the Google and Kaggle 5-day AI Agents course has become one of the largest technical courses ever conducted.
Good news for anyone who missed the live week: it’s all now free to watch at your own pace, and there’s a lot more to it than the hype suggested.
# What it actually is
The course is a continuation of Google and Kaggle’s GenAI Intensive in 2024, which itself attracted Over 140,000 developers and a replay that set a Guinness World Record for the largest virtual AI conference. The 2025 release narrows the focus from generative AI broadly to specific agents, which is a valid call considering how much confusion still surrounds what an agent even is.
The live broadcast lasted five days in November 2025, but everything has since passed turned into a standalone Kaggle guideso you can now work on it on your own schedule. A refreshed version dedicated to vibration coding also became available in June 2026 for those who prefer a cohort experience. Either way, the material is a draw.
# Five days
Every day we combine a technical document with two hands-on programming exercises based on Gemini and the Google Agent Development Kit, so you read a concept and then build it right away.
Day 1 – Introduction to agents: Agent architectures and the basic question of when a task even needs an agent vs. a simpler workflow. You are building your first agent and your first multi-agent system.
Day 2 – Tools and interoperability with MCP: How agents apply tools, how to write custom, and how the Model Context Protocol (MCP) allows agents to communicate with external systems. This day also introduces human approval of long-running operations in the loop, which is the difference between a useful agent and an unsafe one.
Day 3 – Context Engineering: Sessions and Memory: Creating state agents that remember and giving them long-term memory that persists across sessions. This is where most home remedies quietly fall apart, so it’s worth slowing down here.
Day 4 – Agent quality: Capture, track, metrics and evaluate both agent responses and agent tool usage. If you take one day seriously, take this one. Measuring whether an agent is performing is a skill that almost no one teaches and that everyone who puts it into production desperately needs.
Day 5 – Prototype for production: Agent-to-agent protocol and deployment in a managed execution environment such as the Vertex AI Agent Engine. The unsightly last mile between a notebook that impresses your team and a system that real users can apply.
# Who should take it and how to get the most out of it
If you can write in Python and you have called the LLM API, you are ready. The course assumes programming skills, but not agent knowledge, which is exactly the area in which most practitioners currently work.
One piece of advice from observing intensives: don’t look at white papers to rush the code. The assessment and context engineering material from days three and four is where lasting understanding lives, and it’s the part that can’t be retrieved later by copying a working repo. Programming exercises teach you how to build an agent. Newspapers teach you why yours keeps breaking down.
Everything is free now available on Kaggleand the capstone gives you something real to point to when you’re done. One and a half million people found five days for it. Yours are probably worth the same.
Nahla Davies is a programmer and technical writer. Before devoting herself full-time to technical writing, she managed — among other intriguing things — to serve as lead programmer for a 5,000-person experiential branding organization whose clients include: Samsung, Time Warner, Netflix and Sony.