Beyond vibe coding: A controlled and customisable AI software development process

"Beyond vibe coding: A controlled and customisable AI software development process," presented by Philip Langer (EclipseSource), explores how to move from ad-hoc AI coding to structured, reliable workflows for professional software development. This session was recorded at Open Community Experience 2026 (OCX26) in Brussels, Belgium, as part of the Open Community for AI.

This session examines the limits of so-called “vibe coding” in real-world software development and introduces a structured approach to using AI coding assistants effectively.

It highlights why fully autonomous AI coding breaks down in complex environments, especially in systems with high task complexity, mature codebases, and niche technology stacks. In these contexts, missing requirements, architectural constraints, and limited training data lead to unreliable results.

Instead, the session proposes a controlled workflow built around context engineering and iterative refinement. Developers must explicitly provide relevant context, including system knowledge, constraints, and intent, rather than relying on prompts alone.

The approach introduces a structured cycle of defining context, exploring solutions, refining tasks, and validating outputs. AI agents are positioned as tools that amplify developer thinking, not replace it. The session also explains how coding agents operate, focusing on context composition, tool usage, and the importance of managing conversation state.

Finally, it outlines practical techniques to improve outcomes, including breaking down tasks, resetting context when needed, maintaining task-specific working memory, and using persistent project-level context to guide AI behaviour across sessions.

Key topics covered
- limitations of vibe coding in production systems
- AI-assisted software development workflows
- context engineering and task engineering
- defining effective prompts for coding agents
- iterative Define–Review–Refine workflows
- managing LLM context and conversation state
- handling complex and legacy codebases with AI
- role of developer expertise in AI-assisted coding
- balancing automation with control and quality
- structuring reusable AI workflows and prompts

Why this matters
If you let AI run the show, you get fast code and slow problems. This is about using AI without vandalising your architecture.

About OCX26
Open Community Experience 2026 is the Eclipse Foundation’s flagship event, held in Brussels, Belgium. It brings together developers, architects, and industry leaders to explore open source technologies across domains including IoT, AI, automotive, and security, with a focus on practical implementation and collaboration. Learn more at https://www.ocxconf.org/

Chapters
00:00 introduction to AI coding and “vibe coding”
02:24 evolution of AI coding tools
03:16 definition and limits of vibe coding
03:57 why AI coding fails in complex systems
05:18 key constraints: complexity, maturity, stack
06:25 shaping tasks for effective AI usage
07:37 AI as an amplifier of developer thinking
08:27 maintaining code quality with AI
11:02 how AI coding agents work
12:06 understanding context and tool usage
14:16 defining effective prompts
17:32 building context for better results
20:17 structured AI development workflow
22:26 handling imperfect AI outputs
25:06 managing persistent context and memory
27:15 task context and reusable workflows
29:40 optimising agent execution and tools
32:47 conclusion and key takeaways

Видео Beyond vibe coding: A controlled and customisable AI software development process канала Open Community Experience