Foundations
LLMs (Large Language Models)
The models that understand and generate language.
A Large Language Model is a system trained on an enormous amount of text so that it can predict what comes next in a sequence of words. That simple mechanic turns out to be enough to summarise documents, write code, answer questions, translate, classify and reason through problems.
An LLM on its own has no memory of your business, no access to your systems and no ability to take action. It only produces text. Everything else — memory, tools, autonomy — is built around it. That is what the rest of this page describes.
AI Agent
An LLM that can take actions, not just answer.
An AI agent is an LLM wrapped in a loop: it is given a goal, it decides what to do next, it uses a tool, it looks at the result, and it repeats until the goal is met.
The difference from a chatbot is the ability to act. A chatbot tells you what your refund policy says. An agent reads the order, checks the payment provider, issues the refund and emails the customer. Agents are what make AI move from advice to work actually getting done.
Agent Harness
The scaffolding that makes an agent reliable.
A harness is everything around the model that turns a clever text generator into a dependable worker: the instructions it always follows, the tools it is allowed to use, the memory it can read and write, the permissions and guardrails, the logging, and the retry and escalation rules for when something goes wrong.
A model is a component. A harness is the system. Most AI projects fail not because the model is not good enough, but because there is no harness around it — nothing decides what the agent may touch, what it remembers, or what happens when it is unsure.
Plumbing
APIs
How two pieces of software talk to each other.
An API (Application Programming Interface) is a defined way for one system to ask another for something. Your accounting software has an API. So does your CRM, your calendar and your payment provider.
APIs are the reason automation is possible at all. When we say "your tools do not talk to each other", we mean nobody has connected their APIs. When we say "we integrated your systems", we mean we did.
MCPs (Model Context Protocol)
A standard way to give an AI agent access to a tool.
MCP is an open standard for connecting AI models to external tools and data. Instead of writing a bespoke integration for every model and every tool, you expose a tool once as an MCP server and any MCP-aware agent can use it.
In practice this is what lets an agent read your database, search your document store, or post to your project tracker. We build custom MCPs so an agent can safely reach exactly the systems it needs — and nothing more.
Tools
Claude Code
Anthropic’s agentic coding tool, run from the terminal.
Claude Code is a command-line agent from Anthropic that can read a codebase, write and edit files, run commands, and work through a task end to end rather than producing a snippet you paste somewhere.
It is the tool at the centre of our development workflow and of our Claude Code 101 workshop. It is also the clearest everyday example of an agent with a proper harness: it has tools, permissions, memory and a defined loop.
Codex
OpenAI’s coding agent.
Codex is OpenAI’s agentic coding tool, filling a similar role to Claude Code within the OpenAI ecosystem. It reads a repository, plans a change, and edits code.
We use more than one model family on purpose. Different models fail in different ways, so having a second agent review the first one’s work catches problems neither would catch alone.
Buzz
A shared memory layer for a team of agents.
Buzz is the memory system our agents share. Decisions, research, client context, brand assets and lessons learned are written to it once and read back by whichever agent needs them later.
Without a shared memory, every AI session starts from nothing and you spend your time re-explaining context you already gave. With one, the system gets more useful the longer you use it.
Ways of working
SOPs (Standard Operating Procedures)
The written-down version of how your business does something.
An SOP is a documented, repeatable procedure: the steps, in order, with the decisions and the exceptions spelled out.
SOPs matter far more in an AI project than people expect. An agent can only automate a process that someone can describe. In most audits, the first real deliverable is not automation at all — it is finally writing down how the work is actually done.
SKILLs
A packaged capability you can hand to an agent.
A SKILL is a reusable set of instructions that teaches an agent how to perform one specific job the way your organisation wants it performed — your review checklist, your deployment steps, your proposal format.
It is the bridge between an SOP and an agent. You write the procedure once as a SKILL, and every agent on your team executes it consistently from then on. In our workshops, building your first SKILL is the exercise participants leave with.
