Writing
#006 6 min read
Why your agent needs a skill graph, not a prompt chain
Declarative routing is dead. TOML DAGs give agents deterministic, auditable decision paths that prompt chains can't match.
#005 5 min read
MCP beyond hello world: production patterns for tool servers
Model Context Protocol is more than a spec — it's an architecture. Here's how to build tool servers that don't fall over at scale.
#004 5 min read
Agent evals that actually matter: beyond vibe checks
Most agent evals test the wrong thing. Here's a framework for measuring what matters: reliability, cost, and user trust.
#003 5 min read
Context window economics: every token has a cost you're ignoring
Your 200K context window isn't free real estate. Here's how to budget tokens across system prompt, tools, memory, and conversation.
#002 5 min read
Multi-agent handoff patterns that don't lose context
Handing off between agents is where most multi-agent systems fail. Three patterns for keeping context intact across the handoff boundary.
#001 6 min read
The good, the bad, and the ugly of AI agents in 2026
An honest take on where agents actually are: what works, what's hype, and what's genuinely dangerous.