About
coeld
Cedric Oeldorf
Who
I'm Cedric Oeldorf, a researcher and builder working in applied AI. My work focuses on AI systems design — agent architectures, evaluation infrastructure, human-AI interfaces, and the operational problems that arise when deploying language models in real workflows.
Before working, I spent several years building AI-adjacent systems at various scales — from internal tools to production pipelines. I think most about the gap between what AI systems can do in principle and what they reliably do in practice.
What coeld is
coeld is a personal site functioning simultaneously as a research notebook, a guide library, and a lightweight tool surface. It is where I publish thinking that is worth formalizing, document work that is worth sharing, and ship small tools that come out of that process.
The intent is to maintain a coherent public record of a working practice. Not a portfolio optimized for impressiveness, not a newsletter optimized for engagement — just a structured index of work and thought over time.
Lens
I am broadly skeptical of hype in both directions. Language models are genuinely capable of things that matter, and genuinely unreliable in ways that also matter. Most interesting work sits in the tension between those two facts. I try to reason from first principles, verify empirically where possible, and be honest about what I don't know.
Areas of interest
- Agent architectures and multi-step reasoning
- Evaluation methodology and measurement
- Human-AI collaboration patterns
- Interpretability and mechanistic understanding
- AI workflow design and automation
- Information systems and structured knowledge
- Language model behavior and capabilities
- Applied AI in research and knowledge work