Google And Kaggle Reopen Free Five Day AI Agents Course With New Vibe Coding Focus
Google and Kaggle have reopened their free five day AI agents course, now built around vibe coding and a self paced format, after the first edition pulled in more than one point five million learners in a single run.
Highlights:
- Google and Kaggle relaunched their free five day AI Agents Intensive Course after the first run drew 1.5 million learners
- The course now emphasises vibe coding, using natural language instead of complex syntax to build agents
- Content spans five days, from basic agent architecture through to full production deployment
- The programme is now also available as a self paced Kaggle Learn Guide rather than only a fixed cohort
- Completing an optional real world capstone project earns learners an official Kaggle certificate and badge
A course that pulled in one and a half million people in a single run is not the kind of thing a technology company quietly lets fade away. Google brought its free AI agents education programme back for a second run this year, and in doing so, made a fairly deliberate bet about what is currently holding people back from actually building the technology everyone keeps talking about.
Google and Kaggle’s 5 Day AI Agents Intensive Course first launched in November 2025, reaching more than 1.5 million learners in that initial cohort, a genuinely enormous number for what is, at its core, a technical training programme rather than a casual consumer product. A refreshed run followed in June 2026, and the material has since been converted into a self paced Kaggle Learn Guide, meaning anyone can now work through the full curriculum on their own schedule, rather than waiting for the next scheduled cohort window to open.
The most significant change from the first edition to now is a shift the course itself calls vibe coding, a framing where natural language, rather than precise technical syntax, becomes the primary interface for building an AI agent. That shift is not merely cosmetic. It substantially lowers the barrier to entry for people without a formal programming background, opening a course that once assumed real coding fluency to a considerably wider audience of curious, motivated learners.
“Instead of getting bogged down in complex syntax, vibe coding is a paradigm where natural language becomes your primary programming interface,” the course’s own materials explain, describing the goal as letting learners orchestrate genuinely complex agent behaviour using plain English, building what Google’s course materials call 10x agents.
The curriculum itself is structured with a fairly clear, deliberate arc across its five days. Day one introduces the fundamental architecture question underneath every agent project, when a task genuinely needs an autonomous agent versus a simpler, more conventional workflow, before learners build both a first single agent and a first multi agent system. Day three moves into context engineering, the genuinely difficult work of building agents that remember, managing sessions and long term memory so a system retains useful context across multiple interactions rather than starting from a blank slate every time, a stage where, as the course itself notes, most homemade agent projects quietly fall apart. Day four covers agent quality, testing, guardrails, evaluation, and defending against emerging threat vectors specific to autonomous systems. Day five carries the work from prototype to production, covering agent to agent communication protocols and deployment onto a managed runtime like Vertex AI Agent Engine, the unglamorous last mile between a notebook that impresses a small team and a system real users can actually rely on.
Each day pairs a technical whitepaper with two hands on code labs, built using Gemini and Google’s own Agent Development Kit, an open source, model agnostic framework supporting Python, TypeScript, Go, and Java, alongside a daily livestream hosted on Kaggle’s YouTube channel featuring the researchers and engineers who actually built the curriculum.
Promoting the course’s return on LinkedIn, Neil Hoyne, Chief Strategist at Google, helped drive attention to the relaunch, part of a broader company push to position agent building skills as a genuinely accessible, mainstream capability rather than one confined to specialist machine learning teams.
For learners motivated by more than curiosity alone, the course includes an optional real world capstone project, designing, building, and deploying a complete agent system rather than simply following along with guided tutorials. Completing it earns an official Kaggle certificate and badge, with the strongest submissions receiving Kaggle branded merchandise and public recognition from Google itself.
There are a few practical caveats worth naming plainly before treating this as a course genuinely open to absolute beginners. Despite the vibe coding framing, the course still assumes learners can write basic Python and have made at least one call to a large language model API before starting, a meaningfully lower bar than traditional machine learning education, but still a real technical floor rather than zero prior experience. Participants also need a free Kaggle account, which requires phone number verification, and access to Google’s own AI Studio, meaning the course, while genuinely free of any monetary cost, still asks for a small amount of upfront setup and identity verification before anyone can begin.
It is worth applying a fair amount of scrutiny to what a course of this kind is actually optimised to achieve, alongside genuine appreciation for its accessibility. Google is not offering this training purely out of educational generosity, the entire curriculum is built specifically around Google’s own tools, Gemini, Vertex AI, Google AI Studio, and its Agent Development Kit, meaning learners emerge from the course fluent specifically in Google’s ecosystem rather than a genuinely platform neutral understanding of agent architecture. The underlying conceptual knowledge, when does a task need an agent, how do you manage memory and context, how do you evaluate whether an agent actually works, transfers well to other ecosystems including OpenAI’s or Anthropic’s own agent tooling, but anyone hoping to deploy what they learn on a rival platform will need to do meaningful translation work of their own once the course itself ends.
There is also a fair question about how much genuine skill a free, self paced, five day course can actually confer, against the sheer scale of enrollment Google likes to advertise. A million and a half people signing up for a course is not the same as a million and a half people completing it, still less mastering it well enough to build genuinely production ready agent systems on their own, and completion rates for free online courses at this scale have historically run considerably lower than initial enrollment numbers might suggest to a casual observer reading only the headline figure.
None of this undercuts the genuine value of what Google and Kaggle have built here. A free, well structured curriculum covering the full arc from basic agent theory through to production deployment, built by the researchers who actually work on this technology daily, is a genuinely useful resource in a moment when agent building skills are becoming a meaningfully important professional capability across a widening range of industries. Whether this specific course becomes the on ramp that turns curious learners into genuinely capable agent builders, or simply the most widely publicised entry point into a skill people ultimately need considerably more practice to actually master, is the question this relaunch sets up without a single enrollment number, however large, fully answering on its own.
