Welcome to this week's edition of Tying Shoelaces—where we untangle the complexity of modern AI systems.
The Big Picture
Something shifted this week. After months of watching teams struggle with monolithic AI agents, I'm seeing a clear pattern emerge: the future is distributed.
Three announcements caught my attention:
- Anthropic's MCP gets code execution — Finally admitting that tool definitions don't scale
- OpenAI's agent framework — Moving toward multi-agent architectures
- Google's A2A exploration — Agent-to-agent communication is heating up
The industry is converging on what we've been building for months: specialized agents that collaborate.
What I Shipped This Week
We released version 0.9 of the tyingshoelaces platform. Key updates:
- Workflow states — Track content through creation → review → publish
- Multi-platform publishing — One piece of content, five platforms
- Agent analytics — Finally understand what your agents are doing
If you're running AI agents in production, these features will save you hours of debugging.
The Reading List
Three pieces worth your time:
- Why Context Windows Are a Trap — Great breakdown of why bigger isn't always better
- The Economics of AI Agents — Cost analysis that every PM should read
- Rust for AI Infrastructure — Why performance matters at the edges
Quote of the Week
"The best AI systems don't replace human judgment. They amplify it."
That's it for this week. Reply with what you're building—I read every response.
— Edward
Subscribe for weekly insights on AI engineering and agent systems.