Context·Agents·Platform Engineering
10 Best Practices for AI Context Management
Most teams managing AI context are using markdown files. Here's what better looks like.
Craig Tracey ·
Platform Engineering·Context
The Service Catalog Problem Isn't Backstage. It's the Catalog.
Backstage is hard to run and Cortex and OpsLevel made it easier. None of them changed what a catalog is: a human-authored approximation that drifts.
Craig Tracey ·
Agents·Context·Governance·MCP
Why Live Knowledge Graphs Are the Missing Context Layer for Safe Agentic AI in 2026
Support agents fail in production because they lack live context: who owns the account, what they use, what is at risk. Live graphs fix the substrate.
Craig Tracey ·
MCP·Agents
Operating MCP Servers in Production
Getting an MCP server working is not the hard part. Operating one is: who can access what, catching failures early, and enough visibility to improve.
Craig Tracey ·
MCP·Agents
Building MCP Servers That Models Can Actually Use
MCP has moved fast, but implementation quality lags adoption. Tool design and context quality are the two areas where most MCP servers fail.
Craig Tracey ·
AI·Context
Your RAG Passed Every Test and Failed Every User
Eval overfitting is real and underdiagnosed. Most RAG systems have no model of the organization. They have documents about it.
Craig Tracey ·
AI·Agents·Platform Engineering
Building AI Agents: The Fundamentals
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
We Gave LLMs 150 Tools: Here's What Broke.
We benchmarked six LLMs with 25 to 150 MCP tools and measured accuracy loss, latency spikes, and hard API limits. The cheapest model won.
Craig Tracey ·
MCP
MCP vs CLI: You're Asking the Wrong Question
The debate between MCP and CLI for agent tooling misses the point. The real question is which mode you are building for, and where token costs hide.
Craig Tracey ·
Platform Engineering·Agents·Context
Internal Developer Portals vs Context Layers
IDPs were built for humans browsing catalogs. Agents need queryable relationships, real-time state, and cross-system reasoning. IDPs cannot close that gap.
Craig Tracey ·
Context·Agents
The Connective Layer
Andreessen Horowitz published their thesis on why data agents need a context layer: canonical entities, identity resolution, governance. We build this.
Craig Tracey ·
Platform Engineering
Service Catalog vs Live Ontology: Why Static Catalogs Fail
Service catalogs promised a single source of truth for your infrastructure. Why they go stale, what a live ontology gives you, and when each one fits.
Craig Tracey ·