AI Agents Are Not Magic

AI agents might look impressive, but they’re built from simple ideas.

This talk breaks them down into their core parts: an LLM, a set of tools, and a loop that ties everything together. Using a minimal Kotlin example (~100 lines), we explore how these pieces create agent-like behavior.

No magic. Just patterns you can understand and build yourself.

This is a shorter version of another presentation on the same topic that can be found at:
https://speakerdeck.com/scarrilho/building-an-ai-coding-agent-in-100-lines-of-kotlin

The demo code can be found at: Code: https://github.com/scarrilho/code-agent-example/tree/feature/simple-agent
Companion Medium article: https://sergiocarrilho.medium.com/building-an-ai-coding-agent-in-100-lines-of-kotlin-f9af28e6aeb0

Transcript

1. AI AGENTS 101 by: Sergio Carrilho The building blocks in 100 lines of Kotlin @sergiocarrilho
2. Magic? source: https://claude.com/ja-jp/product/claude-code?1e959936_page=1&38c1d113_page=3
3. Nope!
4. Agents are simple
7. LLM alone can’t do much
8. it needs tools...
11. The “loop” makes it agentic
12. Example
13. Show me the code https://github.com/scarrilho/code-agent- example/tree/feature/simple-agent
14. Repo: https://github.com/scarrilho/code-agent-example Agent.kt: The brain build.gradle.kts Tools: The hands libs.versions.toml Agent files Project files
15. DEMO
16. Pattern everywhere
17. https://github.com/karpathy/autoresearch
18. “Next time you see an AI agent… don’t think magic.”
19. “Next time you see an AI agent… don’t think magic.”
20. Thank You! Thank You! sergiocarrilho.com links at: