Please subscribe to our YouTube channel @ https://www.youtube.com/@DevoxxForever
Subscribe to LinkedIn @ https://www.linkedin.com/company/voxxed-days-amsterdam
Follow us on Twitter @ https://twitter.com/voxxedamsterdam

To tell an AI coding agent what to build, you must remove all ambiguity. But removing ambiguity is literally what writing code is. So did the complexity actually disappear, or did it just move?

Working effectively with AI agents turns out to look a lot like being a good engineer taking requirements from a product owner: you don't write the code yourself, but you set constraints, probe for gaps, and iterate fast when something's off.

In this live demo I'll show the techniques that actually close that gap. Things like grilling your own spec before the agent sees it, and using visualization to catch misunderstandings before they become code. I'll do it with pi, the open-source coding agent harness I think every developer should be using right now, and make the case for why a tool you shape to fit your thinking beats one you adapt yourself to. You'll leave with concrete techniques you can use yourself, and a different mental model for working with AI.

Devoxx

A software engineer at Picnic shares two years of experience using AI coding agents, covering practical pitfalls and tips. Key lessons: don't outsource thinking to the agent — stay involved in research and planning to narrow the decision space. Avoid over-specifying (the 'spec paradox' where writing a full spec takes as long as coding it yourself). Give agents higher-level goals rather than micro-tasks, and ask for proof of completion. The talk also introduces 'pi', an open-source LLM terminal harness, and two workflow skills ('grow me' and 'pi interview') that use HTML forms to speed up the idea-to-implementation-plan process.

pi and the Spec Paradox by Max Sumrall