Writing by Jay Zeng
A place for ideas that need more room.
Field reports and essays on agent memory, coding agents, applied AI, platform engineering, and the organizations that build them.
On the desk
Working theses, made shareable.
Evidence-led writing on memory, interfaces, infrastructure, and the organizational choices behind AI-native systems.
Agent memory · Field report
What 1000+ coding agent sessions taught me about LLM memory.
Context is not continuity, transcripts are not memory, and durable state should survive the agent.
Read essay 02Agent memory · Guide
What is agent memory?
Agent memory is durable state that helps an AI agent preserve useful facts, decisions, preferences, and project knowledge across sessions.
Read essay 03Agent memory · Concepts
Agent memory vs. context: why bigger context windows do not create continuity
Context determines what an AI model can see now. Agent memory determines what should remain useful across sessions.
Read essay 04Agent memory · Concepts
Agent memory vs. transcripts: session history is evidence, not memory
Agent transcripts preserve what happened. Useful memory selects and maintains what a future agent should know.
Read essay 05Agents & architecture
The semantic layer agents are missing
Software built for agents needs to expose meaning, intent, and safe action—not merely endpoints.
Read essay 06Agent memory
Memory is infrastructure, not a prompt trick
Useful memory is a lifecycle: capture, retrieval, context budgeting, correction, and recovery.
Read essay 07Engineering organizations
What AI-native engineering actually changes
Coding assistance is the visible edge. The deeper change is how organizations express and move intent.
Read essayFour essays. More only when the idea earns the space. Plus three Agent Memory primers.
Start a conversation