MCP·Agents
Getting an MCP server to work is not the hard part. The hard part is operating one: knowing who can access what, catching failures before users do, and building enough visibility to improve over time.
Craig Tracey ·
MCP·Agents
Model Context Protocol has moved fast, but implementation quality lags behind adoption. This post covers tool design and context quality: the two areas where most MCP servers fail.
Craig Tracey ·
AI·Context
Eval overfitting is real and underdiagnosed. Most RAG systems are built on a flat, static, insider-authored approximation of organizational knowledge. They don't have a model of the organization. They have documents about it.
Craig Tracey ·
AI·Agents·Platform Engineering
Twelve rules for building AI agents that actually work. What agents are, how the agentic loop works, and the mental models that matter.
Craig Tracey ·
Agents·MCP
MCP tool overload is real. We benchmarked six LLMs with 25 to 150 MCP tools and measured accuracy degradation, latency spikes, and hard API limits. The cheapest model won.
Craig Tracey ·
MCP
The debate between MCP and CLI for agent tooling misses the point. The real question is which mode you're building for, and where the actual token costs hide.
Craig Tracey ·
Platform Engineering·Agents·Context
IDPs were built for humans browsing catalogs. AI agents need something different: queryable relationships, real-time state, and cross-system reasoning. Here's why IDPs can't close the gap.
Craig Tracey ·
Context·Agents
Andreessen Horowitz published their thesis on why data agents need a context layer. Canonical entities, identity resolution, tribal knowledge, governance. We've been building exactly this.
Craig Tracey ·
Platform Engineering
Service catalogs promised to be the single source of truth for your infrastructure. Here's why they go stale, what a live ontology provides instead, and when each approach actually makes sense.
Craig Tracey ·
MCP·Agents·Architecture
Giving an LLM 40 tools and hoping it picks the right one is the same mistake dashboards made for humans. Progressive disclosure fixes it, but for agents the mechanism is different.
Craig Tracey ·
Developer Experience
Developer onboarding remains painfully slow despite AI coding tools. The root cause is not documentation. It is that system understanding is trapped in tribal knowledge and disconnected tools.
Craig Tracey ·
Platform Engineering
Blast radius analysis maps the downstream impact of changes before they happen, whether that's an API field, a database column, or a key person leaving the company.
Craig Tracey ·