Google and Kaggle's Free AI Agents Course Returns June 15–19 With a Vibe Coding Focus
Google and Kaggle are running their free five-day AI Agents Intensive course again, June 15–19, and this time the whole thing is built around "vibe coding" — which is either the most honest framing in AI education or the most misleading one, depending on your tolerance for terminology that sounds like it belongs on a conference talk you would quietly leave early.
The original iteration ran in November 2025 and attracted over 1.5 million learners. That number is not a typo. Whatever you think of the curriculum or the vibe-coding premise, 1.5 million people wanted structured agent-building education from a credentialed source badly enough to register for a five-day program. That is a real signal about where the demand actually sits, independent of the discourse about whether vibe coding is the future of software development or a fancy word for prompt engineering with extra steps.
What "Vibe Coding" Actually Means in This Context
The course description says you will learn to "design, build, and deploy robust agent systems" using natural language as the primary interface. The emphasis on production-readiness is new this time around. Google is explicitly trying to move beyond the "here is how you call the API" framing and toward something that looks more like a real development workflow: connecting tools and APIs, handling multi-step reasoning, structuring projects that do more than single-shot Q&A.
Here is the honest version of what that means in practice. Vibe coding, as a workflow, describes describing what you want in natural language and having the AI assemble it from available tools. That is a real workflow. A growing number of solo developers and small teams use it daily to ship things. But it is also a workflow that works best when you already understand the underlying systems well enough to catch the AI's mistakes, wire things together when it cannot, and debug when the output does not match the intent.
The production-readiness claim is the part worth scrutinizing. You will learn how to connect Gemini to external tools, handle multi-step reasoning flows, and structure a project. Whether that translates to production-grade systems depends entirely on what you are building and how rigorously you test it. Agent systems fail in ways that are hard to anticipate before you have experienced them: latency cascades, tool call loops, context window pressure on long conversations, silent failures in chains of reasoning. A five-day course cannot cover all of that. But it can give you the mental model to know where to look when things break.
Why the Course Is Worth Taking Seriously Anyway
The credential and the community are the real value proposition. 1.5 million learners in the first run means there is already a built-in network of people who went through the same curriculum. For developers who are new to agent architectures — or who have been experimenting ad hoc and want a more systematic foundation — that network is worth more than any individual lesson.
Google offering this for free through Kaggle, rather than behind a paywall or inside a proprietary platform, is a deliberate acquisition move. The bet is that developers who learn on Google's stack stay on Google's stack. That is cynical, but it is also correct: free education with a specific tooling bias is one of the most effective ways to shape developer habits long-term. If you come up through Gemini tooling, you are more likely to reach for Gemini when you are building for real.
The course structure — five days of conceptual deep dives plus hands-on examples, culminating in a capstone project — is the right shape for adult professional education. It is not trying to turn you into an expert. It is trying to give you enough structured context to stop fumbling and start iterating. That is a useful bar for a free program, and it is a more honest pitch than "become an AI engineer in five days."
What Google Is Arguing About the Future of Development
The interesting story is not the course itself. It is what Google is implicitly arguing by building an entire free curriculum around "vibe coding" as the organizing principle. They are taking a position: natural-language-first agent building is not a gimmick or a training-wheels mode for people who cannot code. It is the next default interface for software development.
Whether that is right is genuinely debatable. The counterargument is familiar: real software engineering requires precision, testability, and reasoning about state that natural language cannot provide. A vibe-coded agent that assembles tools from natural language descriptions is only as reliable as the AI's ability to translate intent into correct tool selection and sequencing, which is not yet at the level where you can stop understanding what is happening under the hood.
But the counterargument underestimates how much of real development work is actually glue code, context-switching, and assembling already-understood primitives into new shapes. That is the part vibe coding attacks most effectively, and it is a larger fraction of day-to-day work than most senior engineers would like to admit when they are being honest.
The course is real, free, and aimed at production. That combination is worth treating seriously, regardless of your prior position on the terminology. If you have been dismissing vibe coding as discourse noise, this course is Google's argument that it is time to form a more considered opinion — because the tooling is real, the adoption is real, and the gap between "vibe coding produces working code" and "vibe coding produces production-grade systems" is exactly the gap that serious developer education is trying to close.
Registration is open now on the Kaggle competition page. If you are already deep in agent architectures, this is probably review. If you are not, it is probably the most efficient on-ramp you will find that is backed by a major model provider and a community of 1.5 million peers.
Sources: Google Blog, Kaggle