Michal (@tangentstorm) tests real-time voice interaction with Grok while coordinating tasks across a Grok Bot swarm. We also take a deep dive into Lauren Tan's pstack architecture—examining how project-local verification skills, feature maps, and persistent virtual drivers enable high-volume, reliable agentic software development.
Key concepts discussed:
- Real-time voice interaction vs. silent agent prompts
- Running autonomous agent swarms in parallel
- Lauren Tan's Pstack framework and "Poteto Mode"
- Self-verifying agents vs. manual human approval loops
- Feature maps as materialized agent memory
Timestamps:
00:00 - Introduction & Real-Time Voice Testing
01:40 - Desktop Context & Meet the Bot Swarm
03:02 - Prompt Snapshot Testing: Banana vs Habanero
04:19 - Delegating Tasks to Artisimo and Skeletor
07:03 - Pstack by Lauren Tan (@poteto)
08:42 - Porting Arcade Demos with the Swarm
10:05 - The Verification Skill Architecture
12:17 - Feature Maps: Materialized Memory for Swarms
13:18 - Driving Virtual Users with the Command Pattern
15:28 - Managing Multi-Agent Token Budgets & Wrap-Up
