# Tzafrir Rehan

> I write software. These days, mostly tools that let people and AI agents work on the same task without getting in each other’s way.

## Work

### AI Architect at Fiverr’s AI Innovations Lab

_Now_

My paper [Test-Driven AI Agent Definition (TDAD)](https://arxiv.org/abs/2603.08806) treats agent prompts as compiled artifacts. You write behavioral specifications, a coding agent turns them into executable tests, and a second coding agent refines the prompt until the tests pass. The [code and benchmark](https://github.com/f-labs-io/tdad-paper-code) are open.

### CTO at [Be.Live](https://be.live)

Be.Live is the live video startup that revolutionized the live video space.

## Projects

### [Claude Code Mods](https://tzafrir.net/mods.md)

_2026_

Three mods that change how Claude Code itself works: a pane for the tasks only you can do, a box that says what Claude is doing right now, and a pane that keeps Claude’s plan in view.

## Elsewhere

- [GitHub](https://github.com/tzafrir)

## Pages

- [Home](https://tzafrir.net/index.md): Tzafrir Rehan writes software, lately tools for people and AI agents working on the same task.
- [Claude Code Mods](https://tzafrir.net/mods.md): Three Claude Code mods: Human in the loop, What's Agent Doing and Longterm.
- [Human in the loop](https://tzafrir.net/mods/human-in-the-loop.md): A Claude Code mod. Claude hands you the tasks only you can do; they wait in a pane until you answer, and your answer goes back to Claude.
- [Longterm](https://tzafrir.net/mods/longterm.md): A Claude Code mod. Claude keeps its plan in a pane beside the chat, ticks steps off as the work moves, and brings the plan back in the next session.
- [What's Agent Doing](https://tzafrir.net/mods/whats-agent-doing.md): A Claude Code mod. A box above the prompt says what Claude is doing right now, in plain English, and how long it has been at it.

## Full text

- [llms-full.txt](https://tzafrir.net/llms-full.txt): every page on this site in one file