Fidget (opens in a new tab)
A little character that lives on your desktop. It walks the tops of your windows, naps, and lets you pick it up and throw it. Put an AI agent behind it and it comes to life.
Welcome to my little neck of the woods. This is where I write about software systems, architecture, processes, leadership, the paths that do not go as planned, why, and how to make them (hopefully) better. Lately this includes figuring out where AI actually helps engineering work, and where it does not.
I write about things I care about - they are usually unglamorous, and rarely innovative, but hopefully concise, insightful and entertaining enough to read through. I hope there's something here for you to take away and apply in your scenario or line of work. That's it. Welcome to Unhappy Path.
Notes on engineering, systems, and leadership.
The most dangerous thing about vibe coding is not AI writing bad code, it's people building things that should never have been built in the first place.
Every shortcut looks reasonable in the moment. You are just putting another empty roll on the tower. When it falls, it falls on whoever happens to be standing there.
Agentic tools are quietly reshaping which Staff+ archetypes thrive and which are squeezed. Not all Engineers are feeling the pressure the same way.
Coming back from a month away and finding the team shipped anyway is not luck. It is the point of the job.
A few things I've built or contributed to.
A little character that lives on your desktop. It walks the tops of your windows, naps, and lets you pick it up and throw it. Put an AI agent behind it and it comes to life.
Agent skills, prompts, and small tools I still use after they have survived contact with real work.
A Technion-era tool that takes an entangled quantum state and finds the closest separable one. C++ and Eigen under a small web front end. After my old Prof and project guide Yosi Avron (opens in a new tab) told me it’s still useful in 2026 it undergone a revival, facelift, and public deployment (Jan 2026).
Git for data and models: versioning, pipelines, and experiment tracking on top of the repo you already have. I was Director of Engineering at what was then Iterative and is now DataChain, and DVC was our strongest project.
Continuous Machine Learning: CI/CD for ML that trains and evaluates models in GitHub Actions or GitLab CI and reports the results back on the pull request. Built while I was Director of Engineering at Iterative (now DataChain).
A high-performance serverless framework for real-time and data-intensive workloads on Kubernetes. Built at Iguazio, where I was leading the platform group.
An open-source MLOps orchestration framework that takes models from training to production serving. From my Iguazio days, when I was leading the platform group.
Recommended reads, newsletters, and tools I keep coming back to.