Jared Alonzo
ResumeCustomer engineer by day. I'm teaching myself to build agentic systems and writing about what I get wrong along the way.
Writing
How Agent Memory Works
The four types of memory in an AI agent, from the ground up — working, semantic, procedural, and episodic, and why memory is curation, not capacity.
Four steps to cut LLM costs by 60–70%
Measurement infrastructure, prompt caching, a deterministic prefilter, and a model swap. Each one validated against the previous baseline.
How to Be the Person Everyone Trusts Across Teams
Being trusted across functions isn't about being the smartest person in the room. It's about being the most legible one.
What small wins actually do for a client relationship
Client relationships aren't won in the big moments. They're sustained in the small ones — and most people don't figure that out until they need it.
What I didn't know about MCP
Adding career context to the agent looked like a few hours of work. It was, but not the hours I expected — MCP-bridged tools don't support sync invocation, and that one constraint touched every layer of the agent.
Debugging a library you didn't write
A diagnostic story: why a third-party library silently swallowed a 403, why verifying your assumptions saves a few headaches, and what it took to build a working live client from scratch.
Two models, two jobs
The classifier and the narrator have different jobs. They should have different prompts, different temperatures, and—it turned out—different models.
Stateless graphs, stateful consumers
Stateless graphs are easier to test. State has to live somewhere. The consumer is a natural home.
Building a Real-Time Analyst Agent Using LangGraph
A walkthrough of building a queue-driven LangGraph agent that analyzes real-time events in sports.