Yad Konrad
Applied researcher, comedian. Deep interest in meta problem solving, making machines write code that write code.
Recent Notes
View all →Why I Keep Coming Back to Reinforcement Learning
My first self-driving project wasn't RL. What is, why it's strange next to the rest of machine learning, and how the same idea runs from 1988 to the way language models and coding agents get trained now.
- What My $200 Codex Plan Was Actually WorthSeven weeks of Codex logs lined up against my weekly limit, the new Pro tiers, and the 62,500 credits that showed up the night after DevDay: what the pricing change actually means for how I use Codex.
- Every Coding Agent Trend Is Program SynthesisPrompt engineering, context engineering, loop engineering, harness design, evals. If you know the source material, you can predict what gets named next.
- Building StackUnderflow: Local-First Observability for Coding AgentsA local-first toolkit that ingests session logs from 17 coding-agent providers to surface cost analytics, filesystem time-travel, and a session memory that agents can query mid-task. It answers two questions about your coding agents: what did this run actually cost, and what did I already learn here?
- Building Chimera: A Coding Agent Framework, Built by Coding AgentsHow building the same agent three times led me to decompose coding agents into composable primitives, and why I didn't write most of the code
- Sutro Yaro: Agent-Driven Research on Energy-Efficient LearningHow a study group in San Francisco turned coding agents into research agents by going back to 1960s AI problems with modern tools
- Building Mycelium: A Digital Garden Theme for People Who Actually WriteWhy I built a digital garden theme around how I actually think and write
Featured Projects
View all →Novalis
A GPU-rendered terminal built for the way work looks when coding agents do half of it — agents get a cockpit, you keep the controls.
staxtrace
Offline, local-first observability and memory for AI coding agents — cost analytics, time-travel playback, and a memory layer your agents query mid-task.
Pacenote
An open-source Python framework for building coding agents. Pick your provider, pick your tools, pick your loop — then race whole agents against each other on the same task.