Latest · June 2026
Agent from Scratch Part 5: Evaluation
A working agent is not a good agent. Evaluation is what closes the loop between shipping and improving.
AI Engineer at Character.AI — Solving multi-modal agentic problems end-to-end. Writing about agents, LLMs, and building things that matter.
Writing
Hermes Agent: When Your Agent Has Too Many Skills
Real-world lesson from running Hermes at scale: too many skills makes your agent go nuts. Here's the system I built to fix it.
Hermes Agent: Building Real Multi-Agent Support
HermesAgent has a built-in delegate_task tool. I found the problem with it — and built process-isolated sub-agents that actually retain what they learn.
Agent from Scratch Part 3: Skills
Skills are user manuals for your agent's tools. Get them wrong and your agent spends more time confused than working.
Agent from Scratch Part 4: Memory
Memory in agents is just expanding the context window. Here's the simple mental model that makes it practical.