Koders Fest

I tried letting AI build a complete app overnight while I slept, providing the project specifications and tracking task progress as if it were a traditional team. The result was better than I expected. In this talk, I will show you the method I use every day to create complex projects with AI—from prototyping and design to actual development and validation—while avoiding security loopholes. We will explore this through real-world examples from professional app developers to understand why "Vibe Coding" today is much more than just asking an AI to "build me an app."

Vibe coding has changed everything, but most teams are using it the wrong way. What does it really mean to move from “asking AI for code” to managing agents that work for you? This talk explores workflows that genuinely boost productivity, the most common mistakes that turn speed into technical debt, and what sets apart those who use AI as a multiplier from those who simply generate code and hope it works

Vibe coding is here. AI-generated code is everywhere. And many organizations are responding by cutting Junior developer roles — a short-sighted strategy that will backfire. This session challenges that assumption and presents an alternative framework: rather than eliminating Junior roles, evolve them into a critical new function — the expert AI code reviewer. The talk explores why AI-generated code (from Copilot, Claude, Gemini, and others) is still frequently low-quality, incorrect, or incomplete on the first try; why blindly trusting it is a strategic mistake; and how Junior developers, trained properly, can become the human layer that ensures code quality and reduces the senior-level firefighting caused by unreviewed AI output. A real consulting case study underpins the framework. Learning Outcomes: Attendees will understand a concrete workforce strategy for the AI coding era: how to evolve (not cut) Junior talent, what the "AI code reviewer" skill set looks like and how to train it, why this model is more cost-effective than full Senior reliance, and how it future-proofs the organization regardless of how AI coding tools evolve. This is immediately actionable for engineering leads, CTOs, and people managers navigating the AI transition.

Generating code via chat is fine for prototypes, but in real projects it leads to hallucinations and technical debt. Debugging AI now often takes more time than writing code itself. The industry is moving toward Spec-Driven Development (SDD). In this talk, we will explore how to replace vague prompts with structured specifications written in Markdown. You will learn how to define clear contracts to govern AI agents, reduce context rot, and produce verifiable code on the first attempt—stopping the need to rewrite everything from scratch.

In modern software teams, the critical parts of the development lifecycle — code reviews, requirement clarification, and test design — are still manual, inconsistent, and hard to scale. What if these responsibilities could be delegated to a team of AI agents? In this talk, it is presented a real-world system built using n8n and OpenClaw that augments the software development lifecycle through collaborative AI agents. We start from a practical use case: an AI-powered code review agent triggered on Bitbucket pull requests. This agent analyzes code changes, validates them against internal guidelines, and checks alignment with Jira tickets and acceptance criteria. From there, we evolve the system into a multi-agent architecture: - a Product Manager agent that refines and structures requirements, - a Developer agent that performs context-aware code reviews, - a QA agent that generates test cases and identifies edge scenarios. Rather than relying on a single general-purpose LLM, this approach models how real teams work: through specialized roles, structured artifacts, and coordinated workflows. The talk will cover: - how to design agent responsibilities and boundaries, - how to orchestrate agents using workflow automation, - lessons learned from building and running such a system, - and common pitfalls when moving from single-agent to multi-agent architectures. Attendees will leave with practical design patterns to build their own AI-driven development workflows — and a new perspective on what “AI teammates” can actually mean in production systems.