I’m a data scientist on the founding team at Accomplish, building and evaluating self-learning AI workflows. I’m interested in how agents learn from experience, and what happens when they get stuck.
I co-created Boundary-Bench to study coding agents under production-like restrictions. Previously, I was on the founding team at Skyline AI, acquired by JLL in 2021.
| Work | The question behind it |
|---|---|
| Accomplish | Can an agent use what happened on one task to do better on the next? |
| Boundary-Bench | How do security restrictions change coding-agent success, cost, and failure? |
| Skill evaluations | Does a skill actually help agents? |
| Agent memory & model routing | What should an agent remember, and which model belongs at each step? |
I like breaking work into manageable tasks and keeping track of what needs doing. I built Gmail Task Indicator, a Chrome extension that adds blue checkmarks to Gmail threads linked to Google Tasks, so email follow-ups are easier to spot.
- Permission Denied: Policy-Graded Evaluation of Coding Agents in Hardened Environments · arXiv, 2026
Dotan Davidovich, Yair Amar, Hai Rozencwajg, and Or Hiltch. - The Hardening Tax: Policy-Graded Evaluation of Coding Agents under Enterprise Security Constraints · NeurIPS 2026 workshop poster
Yair Amar, Dotan Davidovich, Hai Rozencwajg, Or Hiltch, and Ravid Shwartz-Ziv.
🎬 We Tried OSS Model Routing Tools So You Don’t Have To · With Niv Yungelson, AWS × DevOps.
Based in Tel Aviv. About a decade in data science and machine learning, with a BSc and MSc in Computer Science from Bar-Ilan University. Outside work, my kids and I are inventing games together.
Working on agents, evaluations, or memory? Say hello.



