# Who will find the right agent?

Roei Bar Aviv / AgentBooks

Sharon sits down with Roei Bar Aviv, founder of AgentBooks, to explore how people will find, run, and evaluate AI agents. The conversation moves from childhood curiosity about computers to a practical distinction between completing a task and taking responsibility for an outcome.

## A way to use agents without building everything yourself

Roei introduces AgentBooks as an orchestration platform and a catalog of agents that people can reuse. The idea is to make working with agents accessible beyond the audience that can assemble its own development environment.

The discussion covers running agents in the cloud, sharing or cloning their configurations, and giving people a starting point rather than making each user construct the whole system. The analogy is to a repository of reusable software, but with agents that perform work.

## Why some actions still need a human gate

Roei describes working in contexts where permissions and oversight matter, including regulated industries. An agent's ability to take an action is separate from whether it should be allowed to take it without review.

Sharon and Roei connect that point to accountability. An organization needs someone who owns the decision and its consequences. Automation can change how work is done without removing the human responsibility around it.

## Agentic and autonomous are different promises

The conversation separates an agent carrying out a task from a system that takes continuing ownership and starts work when needed. Those are different levels of delegation, with different expectations about judgment, memory, and oversight.

Roei uses the variety of tools in a kitchen to explain why a collection of agents can be preferable to one tool for everything. Different jobs call for different capabilities and permissions. The separation can make responsibility clearer, rather than merely adding complexity.

## Making the platform understandable to agents too

AgentBooks is also discussed as a service that another agent can understand and use. Roei describes documentation and interfaces designed for that route, so that a person can ask their assistant to help set up and operate the service.

This changes the question of who the product is for. The user remains a person with a goal, but an agent may be the immediate reader of the instructions and the operator of the interface. Clear documentation becomes part of the product experience.

## Finding the right agent, not just another agent

Near the end, Sharon and Roei develop the idea of Agent Fit Analysis and AgentRank: a way to help someone find an agent suited to their task. As the number of available agents grows, discovery and evaluation become problems of their own.

The discussion goes beyond popularity. They consider the quality of an agent's work and the cost of obtaining a useful result. AgentRank is presented as an idea developed during the conversation and an invitation to build publicly, not as an already-delivered search engine.

## The return of the joy of building

Roei's story includes early experiments with computers and the feeling that building had become slow and tedious before AI made it exciting again. That personal thread sits alongside the technical discussion.

The closing question is what to do with that renewed ability. Reusable agents, better discovery, and clearer human gates could help more people build and operate things, provided the system remains understandable and someone stays accountable for the result.

## Topics in the conversation

00:23 What is AgentBooks?
04:07 Permissions and regulated use cases
09:28 An agent-native interface
13:12 Roei's path into building
23:14 What is an agent?
25:41 Why use multiple agents?
32:12 Agentic versus autonomous
44:06 Agent Fit Analysis and AgentRank

Source: https://www.youtube.com/watch?v=QCkm4Fd7Uj0

Edited notes, not a verbatim transcript.

[Join the WhatsApp community](https://chat.whatsapp.com/DgKpG63438y8N7D3pHJ57t)