// engineering blog
Insights on the
agent era
Deep dives on AI infrastructure, agent design patterns, and how we build NexusLayer.
Architecture
Why AI Agents Need Infrastructure, Not Middleware
The pattern of adapting human-facing SaaS tools for agent use is slowing down the ecosystem. Here's what purpose-built agent infrastructure looks like and why it matters.
Coming soon · NexusLayer Engineering
Memory
The Three-Layer Memory Model: Working, Episodic, and Lessons
How AgentBrain structures memory the same way human cognition does — and why the lesson graduation system is the key innovation that makes agents actually improve over time.
Coming soon · AgentBrain Team
LLM Routing
Zero Lock-In: Switching From Claude to Gemini in 30 Seconds
A walkthrough of ModelRouter's provider-agnostic design — how we maintain Anthropic SDK compatibility while routing to any backend, and what that means for your cost strategy.
Coming soon · ModelRouter Team
Observability
The Hidden Cost of Agentic Loops: How WatchGrid Finds Them
LLM cost tracking at the run level reveals patterns that per-call metrics miss entirely. This is how we built WatchGrid's cost attribution model and what it exposed in our own agents.
Coming soon · WatchGrid Team