INSIGHTS
Notes on building production AI agents.
Answer-first writing on what AI agents are, when they beat conventional automation, and the evaluation, observability, and cost discipline that gets one past the demo and into daily production use — from a studio that builds and runs them.
AI agents vs RPA: which do you need?
AI agents and RPA solve different problems. When a reasoning agent beats rule-based RPA, when RPA is the right tool, and how to combine them.
How to choose an AI agent development partner in the UK
Seven criteria for choosing an AI agent development company — production evidence, evaluation discipline, cost engineering, ownership — and the questions that test each one.
How to deploy AI agents in production
A practical guide to taking an AI agent from demo to production — evaluation, guardrails, observability, cost control, and a human-in-the-loop where it counts.
What is an agentic workflow? And when to use one
An agentic workflow lets an AI agent decide the steps to reach a goal, rather than following a fixed script. What it is, how it differs from RPA, and when it fits.
What is an AI agent? A production engineer's definition
What an AI agent is, how it differs from a chatbot and from RPA, what it's made of, and why building one that works in production is the real challenge.
The five RAG architectures worth knowing in 2026 — and when to use each
A plain-English guide to the five RAG patterns that matter in 2026 — Hybrid, GraphRAG, Agentic, Corrective and Multimodal — with a real use case for each and how to choose between them.
AI agent cost optimisation: how to control cost per task at scale
You control an AI agent's running cost at design time, not on the invoice — model routing, context discipline, caching, and loop design set most of it. Treat cost per task as a KPI: budget it before you build, track it in production.
AI agent observability: what to monitor in a production agent
AI agent observability means tracing every step an agent takes in production — token cost, latency, tool-call fidelity, and failure modes — so you can see why it behaved as it did.
An AI agent evaluation framework: what to measure and how a harness works
A practical AI agent evaluation framework measures task success, tool-call fidelity, cost, and latency against real data, run by a repeatable harness — so an agent ships measured, not asserted.
AI agent evaluation: how to evaluate an AI agent
You evaluate an AI agent by measuring it against an outcome agreed up front — task success, tool-call fidelity, cost, and latency — on real data, repeatably. An agent should ship measured, not asserted.