Understanding Agent Tooling
Animal Patrol
What you are about to see: a small Cargo-AI demo that turns a plain-language idea into an agent, a reusable Rust tool, and a shareable demo.
The demo uses animal photos as the concrete task. The agent will inspect the images, produce structured findings, and hand the durable record-keeping work to a Cargo-AI tool. The boundary matters: AI handles judgment, owned tools handle repeatable work.
Journey map
Where this presentation goes
We’ll define what an agent is, then build one around animal photos. The AI judges the images, produces structured findings, and calls a Cargo-AI tool to archive the results. Then we’ll see how Cargo-AI packages that workflow so it can be inspected, shared, and owned end to end.
- 01
What is an agent?
A harness for AI models and tools to accomplish a task.
- 02
What is Cargo-AI?
An agent build tool for JSON definitions, Rust tools, and shareable workflows.
- 03
The Animal Patrol story
Animal photos become structured findings and durable garden records.
- 04
Project setup
Start clean, add guidance, and keep the project inspectable.
- 05
Codex handoff
Open the project in Codex and give it the build request.
- 06
Agent creation
Codex creates the photo-audit agent and associated CSV tool.
- 07
Prepared outputs
Show the run results without making the audience wait.
- 08
Profiles and models
Run the same agent through different model profiles.
- 09
Package and share
Turn the project into something other builders can pull and run.
- 10
Own your tools
Move deterministic complexity out of prompts and into code you control.