Latest Posts
-
Local AI Agent Deployment: An AgentSpace Token Cost Test
Can a local model cut AgentSpace's Claude worker token bill? My test plan, candidate tasks, and what I don't know yet.
-
AI Agent Tool Permissions Sandbox: An Agent Testing Its Own Output
Datasette's app_debug() sandbox trick and whether it solves letting AgentSpace worker agents safely test their own output.
-
AI Agent Permission Scoping: After the OpenAI Lesson
After the news of OpenAI agents acting outside their bounds, how I built permission scoping for AI agents into AgentSpace and ChatFlow.
-
Did Google Index My Claude Shared Chat Links?
After the news that Claude's shared chat links get indexed by Google, here's how I audited my AgentSpace demo links and what I changed.
-
Auto-Publishing Instagram Reels: The App Review Myth and the Real Requirements
I researched the Instagram Graph API for automated Reels publishing: App Review isn't as hard as assumed, cross-posting bans are a myth, and the real blocker is something else entirely.
-
AI Agent Order Automation: What DoorDash's CLI Really Shows
DoorDash's dd-cli is normalizing agents that place real orders. Here's what actually breaks in an agent that turns a WhatsApp message into a Shopify order in ChatFlow.
-
Shippy's Agent Lessons: Why Memory Goes Stale
Agent-building lessons from Ai2's Shippy write-up: why memory goes stale, deterministic CLIs, and what will and won't change in AgentSpace.
-
Claude Code Token Overhead: My Headless Sprint Bill
Claude Code's 33k-token overhead means AgentSpace's headless sprint workers pay the full, unamortized bill every time.
-
AI Clips My Livestream: 3 Hours In, 6 Shorts Out
I let AI cut my livestreams into Shorts: viral moment scoring, vertical reframe, karaoke captions and scheduling. Real cost per stream: $0.34.
-
WhatsApp Sales Bot: Orders, Carts and Shipping on AI
A WhatsApp sales bot that confirms cash-on-delivery orders, runs 3-step cart recovery and sends shipping updates. Plus what it still cannot do.
-
AI Agent Memory: What I Built After Losing 3 Hours
AI agents restart from zero every session. Here is the persistent memory file layout, the checkpoint discipline, and what you should never write to memory.
-
Stop Using One Giant Prompt: How I Run AI as an 8-Agent Team
I run 4 SaaS products solo because I treat AI not as a single assistant, but as an 8-person team with clearly separated roles. Here is the full system.
-
My AI Agent Team Runs Without Me: 3 Lessons From Autopilot
How I run AI agents as a team instead of one assistant: the sprint planning method, lead-agent delegation, and 3 hard lessons from building live on stream.