When writing code with AI, the scariest part is not knowing which step will blow up
DataTalksClub's AI Dev Tools Zoomcamp is free and has 1,757 GitHub stars. Its five modules take AI-assisted development from idea to deployment and monitoring. The focus is not on prompts, but on testing, review, shipping, and auditing. You can join the live cohort or study on your own; the course starts on August 31.
AI can write code fast. The flip side of that speed is uncertainty: can the code actually ship, and will it blow up at 3 a.m.?
DataTalksClub has a free hands-on course, AI Dev Tools Zoomcamp, with 1,757 stars on GitHub. It teaches you to build, test, deploy, and audit software using AI development tools, without losing engineering discipline along the way.

The course repository makes its positioning very clear: this is not a course that only teaches prompts, nor is it a course on training models or building RAG systems. The focus is everyday software development: writing code with coding assistants and agents, running tests, going through CI/CD, deploying, and doing code reviews.
In other words, the emphasis is on safely getting AI-written code into production and then watching it run.
Five modules, from idea to launch
The course is built around one workflow: give AI tools the right context, let them do the right kind of work, review the output, test the results, and ship with guardrails.
- Module 1: AI-native development workflow. Turn a vague product idea into specs and a task list, use
AGENTS.mdto give coding agents persistent context, and then loop through implementation and independent verification with PM/engineer/QA roles. - Module 2: Build and ship a full-stack app. Use an AI assistant to build a frontend prototype from scratch, write an OpenAPI contract, implement a FastAPI backend, add authentication and real-time collaboration, and finally replace temporary storage with SQLite and add tests.
- Module 3: Testing, containerization, and deployment. Use Docker Compose to switch from SQLite to Postgres, add integration tests and Playwright end-to-end tests, deploy to AWS, set up a GitHub Actions pipeline, and release only after tests pass.
- Module 4: DevOps and observability. Separate development and production environments, connect OpenTelemetry to Prometheus, Loki, Tempo, and Grafana, set up actionable alerts, and give the coding agent a read-only responder role for incident investigation.
- Module 5: Reusable skills and specialized sub-agents. Use
SKILL.mdto turn repeated workflows into discoverable skills, define PM/engineer/QA sub-agents, and run tasks in parallel using isolated Git worktrees.
At the end, there is a capstone project on a topic of your choice: frontend, backend, OpenAPI contract, persistence, testing, containers, deployment, and documentation—all the way through.
Who it's for and who should stay away
The course requires basic programming ability—Python, JavaScript, or TypeScript are all fine—comfort with the command line, and a basic understanding of Git and GitHub. No AI tool experience is required, and no GPU is needed.
It is not for people who have never written code, nor for people who want to learn model training, fine-tuning, LangChain, or vector databases. The course draws this boundary itself.
How it works and whether it costs money
It's free. Videos, materials, assignments, and live cohort participation cost nothing.
The 2026 cohort starts on August 31. The so-called live cohort is not a mandatory live class; the lectures are all recorded. What it provides is common deadlines, graded assignments, a leaderboard, peer review, and eligibility for a certificate. If the timing doesn't work for you, you can also study on your own and access the materials anytime, but you won't be eligible for a certificate.
There is a complete playlist on YouTube, a dedicated Slack channel for questions, and announcements on Telegram. The registration link and course platform are both in the repository.
What past participants say
The repository includes a few student testimonials. One freelance data scientist said the course shifted him from “building models” to designing AI-assisted systems, helping him move faster and iterate better. A senior software engineer mentioned that the course taught him to use coding agents effectively, debug problems, and get exposure to MCP, tools, and prompts.
Another data scientist used Antigravity with Gemini and Claude, the Context7 MCP server, TypeScript, Python, Next.js, FastAPI, SQLite, and GitHub Actions to build a Finnish learning website.
What these cases have in common: the toolchain keeps changing, but process and discipline are shared.
One reminder
This is a course, not a tool. You have to practice on your own after finishing it. The 2026 materials in the repository are still being finalized, and some videos, assignments, deadlines, and module details may change before the course starts.
When you write code with AI now, which step are you most afraid of getting stuck on?
Course repository: https://github.com/DataTalksClub/ai-dev-tools-zoomcamp
YouTube playlist: https://www.youtube.com/playlist?list=PL3MmuxUbc_hLuyafXPyhTdbF4s_uNhc43
Slack community: https://datatalks.club/slack.html
发布时间: 2026-11-17 11:26