Engineering
Practical patterns for building autonomous AI systems. How we architect, deploy, and operate AI agents at scale.
Practical patterns for building autonomous AI systems. How we architect, deploy, and operate AI agents at scale.
MCP won the standards war with 97M downloads. But some teams are moving away from it. The case for direct CLI tools over protocol abstraction layers.
When to inject information upfront vs discover via tool calls. A practical framework with real token measurements and the formula that actually matters.
Single-agent tools dominate the market. But enterprise customers want teams. Analysis of architectural patterns, coordination strategies, and when to use each approach.
Why we moved from MCP servers to direct CLI tools. Transparent, auditable, context-efficient agent integrations you can actually understand.
When multiple agents work on the same repo, branches collide. Git worktrees let each agent work in isolation. Here's how we use them for parallel execution.
Claude's context window is finite. Bloated context means slow, expensive, and degraded performance. Here's how we keep agents lean and effective.
