Itamar Friedman is cofounder and CEO of Qodo. His work on code generation and review emphasizes test-driven workflows, organizational context and findings that an engineer can inspect before deciding whether to trust a change.[2][6][5]
Start with The Last Human Code Review: Building Trust in AI-Generated Code →
Education
Links
Owned professional account Official Qodo leadership record Current career biography 
GitHub Cofounder and CEO · Qodo
Career
Friedman worked in R&D at Mellanox and Kaminario and cofounded the robotics startup Dreambots before becoming Visualead’s cofounder and CTO. He later led machine-vision work at Alibaba. In 2022 he and Dedy Kredo founded CodiumAI, subsequently renamed Qodo. This moved his focus toward the tests, context and review processes surrounding generated software.[4][3]
Contributions
Use tests to structure an iterative coding process
AlphaCodium, coauthored with Tal Ridnik and Dedy Kredo, frames code generation as a multi-stage flow. The system uses tests and feedback to improve candidate solutions instead of depending on a single direct prompt. The paper evaluates the approach on competitive-programming problems in CodeContests and reports substantial improvements for its tested setup. Its contribution is an executable workflow for studying code-specific errors and edge cases, with results bounded by the models, tasks and evaluation procedure.[6]
Make review findings traceable to shared knowledge
In his code-review talk, Friedman demonstrates findings linked to the team rules and earlier decisions that support them. The context can include service contracts, past pull requests and incident knowledge outside the immediate diff. A review result can then become a handoff to another agent, while unresolved disagreements return to a human. The mechanism combines independent checking with an inspectable explanation of which organizational constraint a change violates.[5]
Tools, writing & talks
Writing, research & talks
Start here
A demonstration of context-aware review, cited team rules and handoffs between people and agents.
Follow one review finding back to the team rule or prior decision behind it. Software-review experience helps; identify which claims a tool can verify and which disagreements require an engineer’s judgment before attempting automation.
Presenter[5]If AI writes code faster than humans can review it, you are inside the problem rather than ahead of it. Review benchmarks have barely moved across recent model releases, so what separates a review that catches a real contract break from one that asks about error handling is context.
Catalog workOriginal sourceA test-based iterative code-generation study evaluated on CodeContests.
Coauthor with Tal Ridnik and Dedy Kredo[6]