Build agentic software that harnesses the power of AI
An online cohort course where you assemble a working application step by step - from the brain, through the decision system, to the interface. In a few weeks you leave with a production-grade agentic application.
Course language
In the stack
A look at the agent's architecture from the inside - each part is a concrete piece of code.
Tools and inputs: how the agent perceives the world and reaches for data.
State graph + model: where decisions are made.
Graph state: what the agent remembers between steps.
Tool calls: how the agent affects the world.
The agent's character and operating rules.
For engineers who want to build real agentic applications with generative AI - whichever side of the stack they come from. In the AI era what counts is engineering judgment and the ease of working across many languages, not one specific background.
You want to go beyond backend and build agents that make decisions, remember context, and reach for knowledge.
You want to give your app a generative interface and agentic logic, not one more form and fetch.
You want to understand the architecture of agentic systems and their safety boundaries before they reach production.
You want to build a real AI product from scratch - from idea and spec to deployment.
That is no problem. If you have not worked with AI models through an API yet, I will teach you that too. We start from the basics of interacting with models and step by step get to building real agentic applications. And if you already know this, you can quickly move ahead.
You can write code fully by hand whenever you want. At the start, though, you prepare the starter repository for spec-driven development, following a set plan, so the work stays organized from the very first step.
GENREX is three steps. GEN (generate) is when AI generates the code. RE (review) is your review of what it produced. EX (explanation) is asking AI for explanations. All so that you understand what is happening as well as possible, which matters especially when you learn new things with AI. We will prepare a dedicated skill for it.
The method name coincidentally echoes 'Generator Rex', the hero Rex with superpowers. Our superpower is using AI well in software development.
You do not need to be an ML expert or know agentic frameworks. Python and JavaScript basics help, but they are not required - I guide you step by step.
The craft has moved from prompt engineering, through context engineering, to loop engineering - shaping the cycle the agent runs, not a single prompt.
act, observe, evaluate, adjust: the four beats the agent repeats until it reaches the goal, with stop conditions you design.
nodes, conditional edges and state, plus checkpoints and interrupts for stop conditions and human-in-the-loop.
the same loop decides how to assemble the interface around the user's intent: generative UI as the last turn of the cycle.
Step by step you add another layer to your agent - and understand why it works.
LangChain: models, prompts, tools. Your agent's first responses.
LangGraph: state graph, routing, loops and conditions.
State, integrations, tool calls against real data.
CopilotKit + AG-UI: your agent gets a UI in React/Next or in Angular.
LangSmith/Langfuse, self-host FastAPI, deploying your agent.
FastAPI + LangGraph - the canonical LangChain and CoAgents ecosystem.
React/Next + CopilotKit - an agent with a real interface.
Angular + CopilotKit - the same agent, native signals and standalone components.
CoAgents / AG-UI connect the brain to the body - no vendor lock-in.
I have been programming for well over a decade. When ChatGPT arrived I quickly dug into the technical side of the field, and ever since I explore, day to day, solutions that genuinely hold up in production.
An official LangChain Expert, LangChain Certified Agent Engineer and contributor to the LangChain repositories. Day to day a practitioner of agentic engineering and the creator of Ai Board Games. I teach what truly survives contact with production.
You learn to build your agent responsibly. The Vigil Guard guardrails layer is woven into the course.
Filtering the agent's inputs and outputs before they reach the user.
Control over what data the agent sees and what it can reveal.
You add guardrails as another layer of your agent.
(AIDR) to catch prompt injection and protect data before anything goes wrongEveryone learns on the same timeline and supports each other - you ask questions on the community platform.
Build a production-grade agentic application with gen-UI - one that thinks, acts and talks, and assembles its own interface around the user's intent. You learn by building: step by step you create an agentic application with generative UI and unlock new capabilities you'll carry not only into your own projects, but also into the competencies you need at the company where you work.
Schedule
Starts: October 19, 2026
Course language
Modules are weekly, and a new one unlocks every Monday at 0:00. Within a module you learn from text and video lessons, and Softy, an AI agent, guides you through the stages and unlocks each one in turn. You work at your own hours, on a shared cohort schedule, with community access for the course duration.
Yes. The instructor is active on the community platform in your cohort group every day except Sunday - you ask questions and get answers, instead of learning on your own.
You build a working agentic application with a generative interface - from the spec, through the agent graph in LangGraph, to gen-UI with CopilotKit. You leave with a production-ready project, not a demo.
You need the basics of JavaScript or TypeScript and Python, plus a working idea of REST and Git. Prior AI or agent experience is not required - one lesson in module 1 brings you up to speed, but we do not dwell on the basics and move quickly to building. You learn best by doing, because that is the easiest way to understand the concepts.
Yes. React/Next and Angular are equal tracks. Materials, examples and the StarterKit come in both flavours, and the agent layer - LangGraph, FastAPI and the AG-UI bridge - is shared, so either way you learn exactly the same agentic engineering.
Plan for roughly 6 to 12 hours per week. A new module opens every Monday, and you work through it at your own hours - Softy guides you through the stages.
Yes. The course ends with a certificate confirming specific skills that you can add to your LinkedIn profile or CV. Soft 3.0 includes three certifications during its run.
Yes. We issue a VAT invoice for every purchase, so you can expense the course through your company.
Yes. You have 14 days to withdraw from the contract, counted from the course start - the prework period is not included. Within that window we refund the full amount.