GenAI and Agentic AI for Software Engineering
In 9 weeks — from a “Hello World” with an AI assistant to defending your own distributed system of AI agents integrated into the full development lifecycle. 7 workshops, each ending with a working artifact for your portfolio. AI as part of the SDLC, not a separate tool.
overview
Stop prompting. Start building agentic systems.
AI in software development is no longer an experiment — it is a new engineering standard that is already changing how we code, review, test and design systems. AI will not take your job. But engineers who do not integrate it into their process are gradually losing their competitive edge.
There is a gap between “I use Copilot” and “I build agent-driven development”:
- AI as a chat assistant does not scale. You re-explain the context in every session, and the result depends on a lucky prompt rather than on a process the whole team can reproduce.
- Fast generation ≠ a fast team. Without specifications, contract testing and automated review, every generated piece of code is one more round of manual checking.
- One assistant for everything is an anti-pattern. Testing, security audit and documentation are different roles with different requirements for the model, the context and the cost.
This program is about the next step in software engineering: from AI-assisted to agent-driven development. You will learn to build reproducible processes (Context → Model → Prompt, specification-driven development), work with specifications and agents, design systems integrated into the SDLC, raise code quality through automated refactoring, testing and review — and scale software delivery while reducing your team’s cognitive load. As a result, you will start thinking like an AI-augmented engineer, and the speed and quality of your projects will grow regardless of the domain.
what you will learn
Instructor
Oleksii Popov
17+ років в інженерії — від розробника до CTO. Впроваджує ШІ в повний цикл розробки та будує AI-native інженерні команди.
- Ex-CTO у CHI Software — інженерна організація з 800 фахівців
- Ex-VP of Engineering у GenAI.Works — платформа оркестрації ШІ-агентів для спільноти з 12+ млн учасників
- Керував інженерними організаціями на 250–300 FTE у Customertimes і S-PRO
- Solution Architect в EPAM; Head of Software Engineering у fintech-компанії Kreditech (Гамбург)
- AWS Certified Solutions Architect – Professional
У SET University керує програмами GenAI and Agentic AI for Software Engineering та Solution Architecture.
Program
Workshop 1: Introduction and “Hello World” with AI→ Artifact: a Node.js “Hello World” API built with two tools + a comparison of approaches
- Introduction to AI coding assistants (GitHub Copilot, Claude Code, Cursor)
- Comparison of the tools and their use cases
- Setting up the development environment
- Building a simple REST API with AI
- Comparing CLI- and IDE-oriented workflows
- the GitHub Copilot (CLI) workflow
- the Claude Code workflow
- a comparison of the approaches and the results
- A working Node.js app with a “Hello World” API, built twice — with GitHub Copilot (CLI) and with Claude Code
- A written comparison of the two approaches and their results
- A fully configured environment for the rest of the course
Workshop 2: Fundamentals — Context, Model, Prompt Framework→ Artifact: a bank transaction parser with fraud pattern detection
- The Context-Model-Prompt triad
- Best practices for context management
- Choosing models for different types of task
- Prompt engineering fundamentals
- Common mistakes and how to avoid them
- works with several formats (CSV, JSON, XML)
- detects potential fraud patterns
- demonstrates correct context management
- is tested with different models (small vs reasoning models)
- A bank transaction parser: CSV/JSON/XML + fraud pattern detection
- A test on small vs reasoning models — with conclusions on where each model is justified
- Demonstrated correct context management
Workshop 3: Specification-Driven Development→ Artifact: a spec-driven KYC API + your own agent config file
- Principles of specification-driven development
- The structure of a specification template
- Agent configuration files (agents.md, cursorrules, claude.md)
- The API-first approach
- Contract testing and validation
- review and analyse the specification document
- create an agent configuration file
- implement the API with AI on the basis of the specification
- verify the implementation against the specification
- A KYC API implemented by AI strictly to the specification, with a compliance check
- Your own agent config file — a reusable artifact for the team
Workshop 4: Benchmarking and Security, Skills→ Artifact: a model benchmark comparison: quality × performance × cost
- Industry performance benchmarks (SWE-bench, GitHub Copilot leaderboards)
- Model selection strategies
- Cost versus quality trade-offs
- Reasoning models vs fast models
- Performance metrics in real-world scenarios
- Security of AI usage
- Adding skills to AI agents
- Review benchmark results and leaderboards
- Test the same task with different models
- Compare the results and the performance
- Analyse the impact of model usage costs
- A benchmark comparison: one task × different models — results, performance, cost impact
- After the workshop, you choose a model based on data, not habit
Workshop 5: AI Developer Workflows (ADW) — Multi-Agent Systems→ Artifact: 2–3 agents integrated into a fraud-detection system
- The concept of AI Developer Workflows (ADW)
- The Architect–Editor pattern
- Orchestrating multi-agent systems
- Types of agent: tester, security auditor, documentation generator and others
- Integration patterns (side-car, in-process integration, API, service)
- Analyse the code of a fraud detection application
- Choose 2–3 agents to build
- Write the agent specifications
- Justify your choice of agents
- Implement the agents you chose
- Integrate the agents with the fraud detection system
- Demonstrate the interaction between the agents
- Extend the functionality of the system with agents
- 2–3 working agents integrated into a fraud detection system, with specifications and a rationale
- Demonstrated agent interaction — a mini version of your future capstone
Workshop 6: PR Review and Automation→ Artifact: an automated PR review system + a CI/CD integration strategy
- AI-assisted PR review workflows
- Tool comparison: Cursor vs Claude Code vs Codex in code review
- CI/CD integration strategies
- Set up an automated code review system
- Compare the results and the recommendations of the tools
- Write substantive review comments
- A configured automated code review system + a CI/CD integration strategy
- A comparative analysis of Cursor / Claude Code / Codex recommendations
Workshop 7: Model Context Protocol (MCP)→ Artifact: multi-server MCP orchestration over GitHub, AWS, PostgreSQL
- MCP architecture and its benefits
- Official MCP servers (Filesystem, Git, Memory, AWS and others)
- Enterprise MCP servers (GitHub, Azure, PostgreSQL and others)
- Orchestrating multiple servers
- Building your own MCP servers
- Set up several MCP servers (GitHub, AWS, PostgreSQL and others)
- Build a data assistant on top of MCP
- Create workflows that use several servers
- Demonstrate orchestration across servers
- Multi-server MCP orchestration: GitHub, AWS, PostgreSQL, etc.
- An MCP-based data assistant (group work)
who it is for
- Software Engineers (Back/Full-Stack/SDET/Automation QA)
- Tech Leads / development team managers
- Developers who want to grow and master integrating AI into their workflows
- If you do not code yet: basic coding skills and Git/CLI are an entry requirement, not a course outcome
- If you have not worked in a real development cycle: without understanding the SDLC, the practical assignments (spec-driven API, PR review, CI/CD) will not make sense
- If you are looking for a theoretical “what is AI” overview: this is a hands-on program — every workshop ends with a working result
- If you expect “magic prompts” and AI writing everything on its own: the course teaches you to design specifications, verification and orchestration
- No-code audience: the course tools (GitHub Copilot, Claude Code, Cursor, MCP) are for engineers
requirements for participants
Capstone Project presentations
Project description
The program’s finale is not a test or an essay. Applying the knowledge, course materials, tools and frameworks you have gained, you develop a full-fledged distributed system of interconnected AI agents that process tasks within a complete pipeline — on the stack of the whole course: Context → Model → Prompt and specifications (workshops 2–3), benchmark-based model selection (4), multi-agent orchestration (5), automated review (6) and MCP integrations (7). Teams can use the suggested project template or choose their own topic based on their interests.
Successfully preparing and defending the final project is a mandatory condition for receiving the course completion certificate: the certificate confirms that you built and defended a working system, not that you watched the recordings.
The level of systems graduates build — from a graduate’s review of the system she built thanks to the course:
“A pipeline that takes one ticket across several repositories. Eight agents that never talk to each other — not by rule, but because none of them holds a tool that could reach another. The only channel is files. […] tests are written first. On one ticket spanning two repositories that produced 765 lines of tests — four of those files were the first tests that repository had ever had.”Elena Chiperi, program graduate
before → after the program
The difference the program makes is in the approach, not in a single tool:
what stays in your portfolio
Artifact 01 · workshop 1A “Hello World” API with two tools — a Node.js app built with GitHub Copilot (CLI) and Claude Code, with a comparison of approaches
Artifact 02 · workshop 2Bank transaction parser — CSV/JSON/XML, fraud pattern detection, a test on small vs reasoning models
Artifact 03 · workshop 3Spec-driven KYC API — plus your own agent config file and a check of the implementation against the specification
Artifact 04 · workshop 4Model benchmark comparison — one task × several models: results, performance, cost analysis
Artifact 05 · workshop 52–3 agents in a fraud-detection system — specifications, implementation, demonstrated interaction
Artifact 06 · workshop 6Automated PR review system — with a Cursor / Claude Code / Codex comparison and a CI/CD integration strategy
Artifact 07 · workshop 7Multi-server MCP orchestration — a data assistant over GitHub, AWS, PostgreSQL
Finale · capstoneCapstone: a distributed system of interconnected agents — a complete pipeline and a mandatory defense
Reviews
Volodymyr Zubchynskyi
Just wrapped up an incredible course on GenAI and Agentic AI for Software Engineering at SET University!
Instead of just using AI to help me write code, I’m now building the AI agents that handle complex development workflows autonomously.
We are moving way beyond simple autocomplete. Here is how the paradigm is shifting:
🔹 GenAI is the assistant — it waits for a prompt and helps you automate processes.
🔹 Agentic AI is the autonomous coworker — you define a high-level objective, and it orchestrates the necessary tools, runs automated tests, and manages the end-to-end implementation.
Moving from a “prompt engineer” to an “agent architect” is fundamentally redefining the software development lifecycle. 💻✨
Alyona Holovko
Happy to share that I’ve completed “GenAI & Agentic AI for Software Engineering” program at SET University!
The program covered a wide range of topics, from GenAI fundamentals to building Agents and Multi-Agent workflows.
I’m glad I can apply this knowledge in my day-to-day work, and I’m excited to keep exploring and learning.
Thank you so much to Oleksii Popov for the interesting and engaging lectures, for reviewing code, sharing your experience and giving valuable feedback. And thank you to SET University for this opportunity. 🙏
Andrii Hnatiuk
Before the course I mostly used AI as an assistant or a brainstorming tool. Afterwards, I see it much more as an engineering system that can — and should — be orchestrated to build other systems. What the program gave me was not so much knowledge of specific tools, which change extremely fast in the AI field, but a rethinking of approaches and general practices that stay relevant regardless of the particular stack. And those practices turned out to be strikingly similar to the ones software engineering has relied on for years — which strongly resonated with my own view.
Special thanks to Oleksii for the format: a constant dialogue with the audience, hands-on tasks, and the ability to adapt to the group’s pace.
I would also like to note the overall level of organization: everything is transparent, clear, and very well balanced for people who study while working. To me, that is exactly what a good engineering course looks like: what stays with you afterwards is not a set of specific recipes but approaches you can transfer to your own tasks.
Elena Chiperi
Finished GenAI and Agentic AI for Software Engineering. Showing what I built with it, rather than the certificate.
A pipeline that takes one ticket across several repositories. Eight agents that never talk to each other — not by rule, but because none of them holds a tool that could reach another. The only channel is files.
It takes nothing on trust — not the ticket, not its own work:
— every claim in the ticket is checked against the code before anything is written;
— an independent reviewer can reject the implementation, and a stronger model tier rewrites it;
— tests are written first, before the code.
On one ticket spanning two repositories that produced 765 lines of tests — four of those files were the first tests that repository had ever had.
And I measured it against a plain single session, seven times, writing down what I expected before each run. So I know not only that it works, but when not to reach for it.
Thank you, SET University — this is knowledge I will be using, all of it.
FAQ
Can I apply for the program if I do not know Python / JS?
Yes. The course focuses on tools, workflows and principles (Context → Model → Prompt, specifications, MCP, agents) that do not depend on a specific programming language. Python/JavaScript are used only for demo examples.
What is the language of instruction, and why English B1?
Classes are taught in Ukrainian; course materials are in English. B1 is enough to work confidently with tool documentation, specifications and benchmarks used in the original.
What do I need to do to get the certificate, and what exactly will I receive?
Successfully preparing and defending the capstone project is mandatory. You will receive a SET University certificate, 1.5 ECTS credits and a portfolio of working artifacts from all seven workshops.
Can I combine the program with work?
The program lasts 9 weeks: online workshops plus a practical assignment after each one — hands-on practice, not multiple-choice tests. On how the learning is organized, graduate Andrii Hnatiuk writes: “everything is transparent, clear, and very well balanced for people who study while working”.
Can I do the capstone on my own topic rather than a template?
Yes. Teams can use the suggested project template or choose their own topic based on their interests — the only requirement is a full-fledged distributed system of interconnected agents within a complete pipeline.
Which tools are used in the course?
In the course we work with GitHub Copilot, Claude Code and Cursor (in the PR review workshop, also Codex for comparison), as well as MCP servers (GitHub, AWS, PostgreSQL, etc.).
Can I pay in installments?
Yes, installment payment is available. The program manager will send you the installment details after you apply.
Learn more about the SET University program
