Insights
Loading insights
Voxyz AI
BetaEssays behind the threads: stack, failures, swarm patterns, and lessons from building with five agents and one founder in the loop.
Featured experiment
My agents had memory but didn't use it. Process beats reminders. 30% baseline to a 90% live variant across formal suites and follow-up optimization rounds.
Live tracker
Builders' demand pipeline. The ideas we've validated, the experiments shipping, and the bets that didn't pan out. Updated weekly.
Newest first
6 stories
Ten hard-won OpenClaw lessons about tools, context limits, token waste, model choice, and the mistakes that cost money after install day.
Read article
What Voxyz AI learned from its first reorg: remove fake jobs, collapse redundant work, and design every agent around a downstream consumer.
Read articleOlder writing
These pieces still have value, but they reflect earlier naming, older system phases, or historical paths outside the current five-agent product story.
3 older pieces
A practical look at the $8 VPS, cron jobs, monitoring, and runtime boundaries that keep an always-on AI team moving overnight.
Role cards, hard bans, relationship drift, RPG stats, and 3D avatars turned six agents from generic prompts into memorable teammates.
A practical build guide for six AI agents with shared memory, queues, reviews, and a visible operating loop that runs a real company.
