Back to index
Other5 daily units of whitepaper, podcast, codelabs and a recorded livestream, roughly 15 to 25 hours self-paced·Free; the optional Day 5 deployment labs may need a billing-enabled Google Cloud account

5-Day AI Agents: Intensive Vibe Coding Course with Google (Kaggle)

4.2

Five strong whitepapers wrapped in what is partly an onboarding course for Google Antigravity. Worth doing for the reading and the security day, as long as you know the tooling is one vendor's stack.

What We Liked

  • Free and fully archived, so you can do it at your own pace instead of surviving a live week
  • The whitepapers are the best part and stand alone: protocols, Agent Skills, security and spec-driven development
  • Day 4 on agent security and evaluation covers ground most agent courses skip entirely
  • Day 5 is honest that vibe-coded prototypes are fragile and teaches specifications as the fix
  • Recorded livestreams with the Google engineers who wrote the labs

What Could Be Better

  • Every codelab runs on Antigravity, ADK and Google Cloud, so the hands-on skills transfer less than the ideas do
  • The production labs on Day 5 are optional because they may need a cloud billing account, which is exactly where most people stop
  • No certificate or assessment on the self-paced guide
  • Assumes you can already code; the vibe coding label makes it sound more beginner friendly than it is
  • The live capstone window has closed, so self-paced learners get no project feedback

Detailed review

Google and Kaggle have now run this intensive format several times. The generative AI version came first, the agents version followed in November 2025 and Google says it reached over 1.5 million learners, and this June 2026 edition was rebuilt around vibe coding and drew a bit over 353,000 registrations. That is a big drop, and I think it says more about fatigue with the format than about the content, which is better than the hype around it suggests. The structure is the same as before. Each day has a whitepaper written by Google engineers, a podcast summary of it, two codelabs, and a recorded livestream.

Day 1 covers the shift from writing syntax to directing agents and gets you set up on Antigravity 2.0, its IDE and its CLI. Day 2 is tools and interoperability: MCP, agent to agent communication, generative UI, and the newer payment and commerce protocols. Day 3 is Agent Skills, the pattern of packaging instructions in a folder around a SKILL.md file and loading them only when needed. Day 4 is security and evaluation.

Day 5 is spec-driven development and deploying to Google Cloud. The whitepapers are why I rate this well. The Day 3 paper is the clearest explanation I have read of why skills beat giant system prompts, and it applies to any agent framework. Day 4 deals with sandboxing, hallucinated package names that attackers register, human in the loop approval and trajectory evaluation, and the lab has you build an expense approval agent with a manual review step. Very few free courses touch this, and it is the material that matters once an agent can run code on your machine. I also like that the course argues against its own title by the end.

After four days of vibe coding, Day 5 tells you to treat generated code as disposable and written specifications as the source of truth. I agree with that, and it is a more grown-up position than most vibe coding content takes. My reservations are about the tooling. Every lab uses Antigravity, the Agent Development Kit and Cloud Run. If you work in Claude Code, Cursor or anything else, you will be installing a new IDE to follow along, and a fair amount of lab time goes on learning that product.

The Day 5 deployment labs are marked optional because they may need a billing account, so the part about getting to production is the part most people will only read. There is also no certificate on the Learn guide and the capstone has closed. Over 6,000 projects were submitted during the live run, and that feedback loop is gone for anyone starting now. I would still send a working developer here before most paid agent courses. Just read it as Google's view of the field, and build your own project in whatever tools you already use.

[ final ]

The verdict.

Read all five whitepapers, do the Day 3 and Day 4 labs, and skim the rest unless you plan to build on Google's stack. It is a better course about how agent development is changing than it is a neutral course on how to build agents.