Writing, archived.
LinkedIn writing that reaches 100,000+ people a year, plus 10 peer-reviewed papers — on agents, RAG, LLM systems, and the shift from automation to agentic AI. The best-performing pieces, by real engagement — search or filter to find one.
articles
Six parts, one series — first principles to production readiness. Every part linked below, plus where to follow along live.
Tokens, attention, and next-token prediction — the mental model that makes everything else about agents make sense.
Capability gets an agent built. Identity, scope, and a registry entry are what get it approved to run.
Claude.ai for thinking, Claude Code for building, the API for shipping — the practical split, not the marketing one.
No app, no dashboard — just a message thread that reads your accounts, drafts the things you'd forget, and keeps a plan instead of a list.
linkedin posts
Python + the OpenAI API in three steps — the on-ramp for the rest of the series, before a single agent gets built.
LangGraph vs. crew.ai side by side — starting with a single-agent setup and scaling to multi-agent + MCP as use cases grow.
Tools for action, memory for context — why combining both is what turns a static chatbot into something that behaves like an assistant.
A nine-point checklist for the part nobody's demo covers: taking an agent from prototype to a monitored, secured production deployment.
Why single-purpose agents beat generic assistants — precision, speed, and consistency, with real examples from GitHub and CrewAI.
Single agent to multi-agent system, event-driven automation, production deployment, and the observability that makes scaling safe.
A survey of the architectural patterns agentic systems are actually being built on right now.
Retrieval-augmented generation end to end — from vector databases to the retrieval layer that actually determines answer quality.
A layer-by-layer breakdown of what actually makes an agent effective — interface, reasoning, memory, tools, and the layers most builders skip.
Guiding principles for making agents smarter and safer as they move from novelty to infrastructure.
Twelve pain points that show up the moment an agent leaves the demo and has to survive contact with production.
Untangling five terms that get used interchangeably but mean genuinely different things.
A first-principles answer for anyone still fuzzy on what separates an "agent" from a chatbot.
The real skill is understanding how intelligence can act, think, and make decisions autonomously — a primer for developers, founders, and enthusiasts.
Tools and frameworks change yearly — these are the timeless algorithms behind everything from recommendation systems to GPT-style models.
How you chunk text before embedding it quietly determines how good your retrieval actually is.
Practical, narrow agents for hiring, knowledge work, and operations — the kind that ship, not the kind that stay a demo.
Why the format you feed an LLM quietly drives your API bill — and what TOON changes about that math.
A field guide to the frameworks powering the shift from prompting models to orchestrating agents that plan, reason, and execute.
A one-stop map of the tools and platforms making up the fast-expanding LLM ecosystem.
As agents move from experiments to production, the attack surface grows — credential theft, prompt injection, and the vulnerabilities teams underestimate.