PI CONVERSATIONS
From harnesses and workflows to code quality and changing models. Listen to what Mario and Armin are thinking about.
27:26· Monday Meditations
Pi Coding Agent’s session model suits a person working with an agent. Pi Durable targets long-running, multiplayer agents with parallel branches that can suspend and resume across different execution environments.
Watch conversation →
24:26· Monday Meditations
They observe newer models preferring Bash and Python over dedicated edit tools, affecting error feedback, diff presentation and the user’s visibility into execution.
Watch conversation →
6:14· Sunday Meditations
Armin reflects on eighteen months of agent tools: model capabilities have improved, while trust, quality and understanding remain difficult engineering problems.
Watch conversation →
13:21· Sunday Meditations
They argue that excessive generated code consumes limited context and compounds quality problems. Critical infrastructure, interfaces and types still need human control even when not every implementation deserves line-by-line attention.
Watch conversation →
2:20· Sunday Meditations
Different working habits motivate a customizable harness. Pi extensions can be written and reloaded in the current session, with immediate feedback guiding the next revision.
Watch conversation →