Notes on intelligence for AI agents.
Deep dives on memory layers, company intelligence, and the intelligence layer AI agents are missing.
Intelligence Layers in AI Agents: The Architecture That Separates Production Systems from Demos
Most AI agents have no intelligence layer. They have a retrieval call. Here is what an explicit intelligence layer does, the three architectures that exist, and how to tell which one you actually have.
8 Security Loopholes in Agentic Infrastructure Nobody Is Patching
Gartner predicts 25% of enterprise breaches by 2028 traced to AI agent abuse. The attack surfaces are real and already being exploited. Here is what the architecture needs to close them.
Vertical AI Agents: Why YC Says This Category Is 10x Bigger Than SaaS
SaaS replaced manual work with software. Vertical AI agents replace the software and the human operating it. The 10x claim is architectural, not hype, and the context layer is the deciding factor.
What Top Accelerators Are Saying About AI Agent Context, The Signal Most Builders Are Missing
YC, Sequoia, a16z, and Madrona all said the same thing in 2026: the model is solved, the context and memory layer is not. Here is the coherent message, synthesized.
The Best LLM for AI Agents in 2026: Benchmarks, Tool Use, Pricing
Six frontier models within 1.3% on SWE-bench. A 25x pricing gap. The honest answer is model routing - and Harness Engineering still matters more than the model.
Memory in Robotics: What Transfers to AI Agents, and What Doesn’t
40 years of robotics memory research. Episodic memory and continual learning transfer. SLAM and motion planning don’t. Here’s the line.
Why the Context Graph Is the Future of AI Memory
Every serious AI memory system converged on a graph layer in 2025-2026. Here is why vectors cannot replace graphs, what a production context graph actually contains, and what the five-axis scoring model does.
The Explosive Growth of Open-Source Memory Layers (And Why It Matters)
Mem0 hit 41K stars and 186M API calls/quarter. Graphiti hit 24K stars. The memory-layer market just became real infrastructure.
Where YC AI Agent Startups Are Failing: The Honest Post-Mortem
Gartner predicts 40%+ of agentic AI projects canceled by 2027. The failures are architectural, not model-driven. Here are the seven failure modes playing out right now.
Why Harness Engineering Is the Future of AI Agent Development
Six frontier models within 1.3% on SWE-bench. Model picking is dying. The next decade of AI engineering is harness engineering.
How YC Startups Are Actually Shipping AI Agents in 2026
60% AI companies, five architectural patterns that separate shippers from demo-builders, and what YC's own RFS says is still missing.
Databases for AI Agents: The Honest Architecture Guide for 2026
Five capabilities, five database types, three production stacks. The integration tax of doing it yourself - and how to avoid it.
How Garry Tan Shipped GBrain in 12 Days, And What It Reveals About the Future of AI Agent Memory
A working multi-agent system in 12 days, with the context layer as the longest phase. What YC's CEO built tells you everything about where AI agent infrastructure is going.
Harness Engineering: The Discipline That’s Eating Prompt Engineering
Harness Engineering: OpenAI named it, Martin Fowler formalized it, Anthropic and Microsoft adopted it. The two-part framework - guides + sensors - and what to build first.
How Karpathy's LLM Wiki Changes How Engineers Think About AI Memory
The compiler analogy that inverts RAG, the four operations that make a knowledge base compound over time, and what it means for organizational-scale AI memory.
Context Engineering: The Tactical Playbook (2026 Patterns)
Six production patterns from Anthropic, Cognition, Manus, Windsurf, and Microsoft that separate teams that ship reliable agents from teams that don’t.
How MemPalace Hit #1 on GitHub in 48 Hours, And What Builders Actually Learned
The Milla Jovovich repo, 41,200 stars, a benchmark controversy, and what the viral moment reveals about the enormous unmet demand for local AI memory.
The 7 Types of AI Agents: A Clean Taxonomy for 2026
Five academic types plus two architectural patterns - and the decision matrix that tells you which one your use case actually needs.
Self-Evolving AI Agents: What They Are, Who’s Building Them, and Why It Matters
Memory remembers facts. A learning loop changes the procedure. The four components that separate "has memory" from "actually learns."
OpenClaw vs Hermes Agent: The Honest 2026 Comparison
250K-star incumbent vs 95K-star insurgent. The architectural difference, the security caveat nobody talks about, and the decision matrix.
The Growth of AI Agents: Adoption, Revenue, and Where the Money Is Going
$10.91B in 2026, 51% of enterprises in production, vertical agents at 62.7% CAGR - and where the next infrastructure dollars are landing.
Multi-Agent Systems in Production: How They Actually Work in 2026
Four production patterns, the 20x cost reality, and the shared-context primitive that turns parallel agents from chaos into a team.
Orchestrator vs AI Agent: The Architectural Distinction Every Builder Gets Wrong
An agent is one execution unit. An orchestrator coordinates many. Mixing them up costs months. Here’s the clean line and the decision matrix.
The Invisible Wall of AI Agents: Why 88% Fail in Production (And How to Get Past It)
Frontier models are within 1.3% of each other on SWE-bench. The 20-30 point gap that decides whether agents ship comes from the harness, the memory, the evaluation - not the model.
The Future of AI Agents: Market, Architecture, and What’s Actually Going to Happen Through 2030
$7.63B to $50B+ by 2030, 40% of enterprise apps with agents by end of 2026, and Gartner’s quiet warning that 40%+ of agentic projects will be canceled before they pay off.
Context Engineering: The Discipline That Replaced Prompt Engineering
Why prompt engineering stopped being enough, the four core strategies (write - select - compress - isolate), and the difference between memory and context.
Open-Source Memory Layers for AI Agents: The Complete 2026 Comparison
Mem0 vs Zep vs Letta vs Graphiti vs Cognee vs Supermemory - the architecture differences, the benchmark numbers, and how to actually pick one.
What YC Founders Are Actually Saying About AI Agent Memory (2025-2026)
Mem0's $24M, Garry Tan’s gstack, the MemPalace credibility incident, and what the YC ecosystem actually believes about agent memory in 2026.
What Andrej Karpathy Is Saying About Memory, Context, and Why AI Agents Don’t Work Yet
Four quotes that reshaped the agent-infrastructure conversation, the LLM Wiki idea, and what context engineering actually means for builders in 2026.
Why Vector Databases Aren’t Enough for AI Agents in 2026
Vector databases solve similarity search. They do not solve Memory Layer requirements. Five failure modes, the architecture that’s actually winning, and why coding agents skip vector retrieval entirely.