AI & Agentic AI
Engineering Intelligence Beyond the Prompt
Exploring the architecture, engineering patterns and governance behind Generative AI, RAG and Agentic systems — from grounded knowledge retrieval to controlled autonomous action. A dedicated collection of my writing on RAG, AI architecture, agentic systems and governed autonomy.
From RAG to Agentic AI: How Enterprise AI Architecture Is Evolving
Enterprise AI is moving beyond simple prompt-response applications. Explore how RAG, tools, reasoning, state and governance progressively change the architecture of intelligent systems.
Read Article →AI Reading Path
Start with RAG. Each piece builds on the one before it.
RAG Explained: A Beginner’s Guide to Retrieval-Augmented Generation
How retrieval-augmented generation works, from documents and embeddings to a grounded answer, and when it is and isn’t the right tool.
Why firstIt introduces the vocabulary the rest of the path builds on.What Makes a RAG System Production-Ready?
The difference between a convincing RAG demo and a production-oriented system: retrieval quality, chunking, embeddings, security, observability, caching, evaluation and operations.
Why nextIt separates a working demo from a system you could operate.RAG vs Agentic AI: When Do You Need Which?
The architectural differences between RAG and agentic systems, and why they are not competing technologies: an agent can use RAG as one of its capabilities.
Why nextIt shows when retrieval is enough and when an agent is justified.Anatomy of an Enterprise AI Agent
The roles of LLM reasoning, tools, memory and state, orchestration, enterprise knowledge, APIs, observability and governance, and why the LLM is only one component of an agent.
Why nextIt covers what an agent needs around the model.Why AI Agents Need Authority Boundaries
The difference between an agent that can propose an action and one that is authorised to execute it, and how risk assessment, deterministic policy, human approval and auditability define the boundary.
Why nextIt separates an agent’s proposal from permission to act.Human-in-the-Loop: Where Should AI Autonomy Stop?
Risk-based autonomy, and why human approval belongs at meaningful authority boundaries rather than around every step an agent takes.
Why nextIt shows where human approval belongs, and where it only adds friction.From RAG to Agentic AI: How Enterprise AI Architecture Is Evolving
Enterprise AI is moving beyond simple prompt-response applications. Explore how RAG, tools, reasoning, state and governance progressively change the architecture of intelligent systems.
Why lastIt puts the whole progression in one view.From Ideas to Engineering
I explore these concepts not only through writing, but by building systems that put the architectural patterns into practice.
Useful autonomy is bounded autonomy.
Deciding what a system may propose, what it is permitted to do and who approves it is engineering judgement, and it belongs in the design from the start.