Joel Sauvain: essays on agentic AI, verification, and building
Hi, I'm Joel, an engineer in Zug, Switzerland, building AI systems that actually ship. These essays are my working notes from the edge of agentic AI: memory, verification, autonomy, and what's left for us to decide.
-
The Autonomy Asymptote
Full autonomy in software engineering: why autonomy lives in the feedback loop rather than the model, why each nine costs ten times the last, and why coding is the first profession to approach the ceiling.
-
Smart Is Cheap. Verified Is Valuable.
Why AI agent verification beats model intelligence: compounding error math, deterministic checks vs LLM judges, and verification as enforcement, not dashboards.
-
Ten Years Out: The Last Decade of Execution Jobs
A signed opinion on 2036: the country of geniuses arrives, intelligence grows hands, execution work disappears, and the ending gets chosen by a politics that is ten years behind.
-
Your Company Has a Brain. It Just Can't Remember Anything.
Why RAG is not memory, what human memory teaches us about AI memory management, and how to build a company brain that consolidates, distills, and forgets.
Next essay by email
About. I've been writing software since I was fourteen and never really stopped: fintech and commodity trading, construction sites in augmented reality, even software for lawyers, and the identity systems and firewalls that guard the rest. These days I build AI-powered products: agentic systems and AI-assisted engineering, where the interesting problem is rarely getting an agent to produce something and almost always proving that it works. On my own time I built acted.app (live on the App Store and Google Play, shipped solo), and a friend and I are building plectin.ai. Along the way I finished an MSc in entrepreneurship at UCL. I write down what I learn here. Say hi on LinkedIn.