Episode 10 · Ben Tossell · October 15, 2026 · 51 min
YouTube · Spotify · Apple Podcasts
About this episode
Ben Tossell thinks of his personal AI agent as a folder: context he can open, inspect, edit, and take to another model. When its memory started holding on to too much old information, he pared it back.
In this outdoor conversation with Michael Galpert, Ben explains how he makes AI useful through small experiments. They discuss the animated mascot on his personal site, learning coding concepts visually, and an email client he built before realizing that labels and archiving were what he actually wanted.
What you’ll learn
- Learning by playing and asking basic questions
- Building prototypes you can throw away
- Making technical vocabulary easier to understand
- Steering coding agents as the work develops
- Keeping personal context portable and inspectable
- Deciding what an agent should remember or forget
- Starting instructions with one line and adding only what is needed
- Building small tools for yourself without turning every idea into a company
Topics and resources
- Ben personal site
- Ben's Bites
- Josh Puckett
- Matt Pocock dictionary
- Index
- Pi coding agent
- Factory / Droid
- Codex
- BB
- Claude / Opus
- Claude Code
- ChatGPT
- GLM / Z.ai
- OpenRouter
- Latent Space / OpenRouter episode
- Pragmatic Engineer / Matt Pocock
- Matt Pocock skills
- ASD Simplified Technical English
- Muse
- Dots
- Grok Bot
- Karpathy LLM Wiki
- Instinct
- Google Drive
- Makerpad background
- Superhuman
- Gmail
- Google Workspace CLI
- Cognition
- OpenAI Sol models
- OpenAI Astra
- AGENTS.md
- Model Context Protocol