The Agentic Pipeline
Learn agentic engineering by building a real autonomous AI system, end to end.
Most AI courses sit at one of two extremes: narrow how-to tutorials, or high-level hype about what AI "will" do. This one sits where the actual work happens. You build a complete autonomous pipeline from scratch, and along the way you learn the engineering judgment that separates a system that demos well from one that actually runs every day without you.
The pipeline you build is a real one. It turns a raw news feed into a polished startup-idea newsletter: ingestion, triage, signal extraction, idea synthesis, deep dive, writing, and as a bonus, a podcast version of each edition. It's the same architecture that powers the GammaVibe Daily Idea newsletter, which has published autonomously every day with no human in the loop for the daily run.
What you actually learn
The headline skill is agentic engineering: directing a coding agent to build real software while you make the decisions that matter. This is normal practice now, but doing it well, knowing what to delegate, how to evaluate what comes back, when to push for a different approach, and how to end up with a system you can maintain, is a skill in itself. That skill is the real subject of the course. The pipeline is how you learn it.
You'll work through the build the way you'd actually do it in practice: ingestion, then layering in the refinements that make a pipeline robust, like vector similarity scoring, image variety, error handling, and retry logic. There's a module on generating the audio overview, and a wrap-up that ties the architecture together. Lessons are kept short and focused, so you work through a stage, build it yourself, and move on.
On the technical side, you'll come away comfortable with:
- Resumable, debuggable pipelines
- AI agents with structured output
- Choosing the right AI models for each task
- Embeddings and vector search
- Using synthetic data to tune ranking functions
- The production trade-offs that decide whether a system survives contact with real data
What you build
By the end you have a working autonomous pipeline that produces a finished newsletter edition: the written deep dive, a header image, and a podcast-style audio version. You'll also have the reference implementation as a public repository, so you can compare your work against a known-good build at every stage, and the judgment to adapt the whole thing to your own purpose.
Who it's for
This course is built for senior engineers and independent builders who want to work with agentic systems at the level of architecture and engineering decisions, not surface-level tutorials. If you ship software, you'll feel at home.
You need working knowledge of Python and APIs. No machine learning background required, there's no model training, no math prerequisites. It's about engineering with AI, not building AI from the ground up.
It's also tool-agnostic. The build uses Claude Code, but the principles apply to whatever coding agent you prefer. You'll need your coding agent, an AI model API, and a news API, which charge modest usage fees as you go.
Meet your instructor
I'm Mirko Froehlich. I spent over a decade shipping products as a software engineer, most recently as a Senior Engineering Manager at Google. Last fall I left to focus on my own projects, and GammaVibe is where I build in public, sharing what I learn about agentic engineering.
This is my first course. It grew directly out of a technical breakdown I published of the GammaVibe pipeline, which generated far more interest than anything else I'd shared, with most of the questions some version of "how do I build this myself?" So I built the answer. It was a lot of work, and a refreshing kind of work, and I'm proud of how it turned out.
If you're not sure whether it's the right fit, you're welcome to reach out and ask.
Not ready to dive in?
Start with the free engineering blueprint. It includes the full architecture diagram, the tech stack breakdown with rationale, starter files, and initial prompts, enough to build your first pipeline step today and get a feel for the approach.