The programmes people are actually running.
Where each programme stalls, and how we would address it. Every guide says above the fold whether this is work we have delivered or an approach we would be taking for the first time.
- patterns, each with its evidence basis stated on the page
- 12
- written from work we have delivered
- 4
- method we would bring, said so up front
- 8
- carry a sources list you can open and check
- 12
Two labels, and we do not blur them
Most firms publish implementation guides for everything and let you assume delivery behind all of it. These are split by what is actually behind them.
Backed by work we have actually done, on real engagements. The guide is written from that work rather than from vendor documentation, and it says plainly where the delivery stopped.
Jump to theseWe have not run this at enterprise scale. The guide is the method we would bring, and it names exactly what we are drawing on instead, at the top of the page rather than in a footnote.
Jump to theseCommon in regulated financial servicesmarks the three UK insurers and banks ask us about most. What each regime requires of an agent is written up on agentic transformation in financial services.
If you arrived with a sentence rather than a category
Each of these is answered in full inside one of the guides, usually in the questions section at the foot of it.
- How do we stop our AI agent hallucinating?
- Do we need to fine-tune, or use RAG?
- What is the ROI of an agentic AI programme?
- How do we stop retrieval leaking documents to the wrong people?
- Should we adopt MCP, or is function calling enough?
- What does an evaluation harness actually consist of?
- What does it cost to run, not to build?
- How do we stop an agent taking an action it cannot undo?
- Does a bigger context window remove the need for RAG?
- Our pilot worked and never shipped. What now?
- When do the EU AI Act high-risk obligations actually apply?
- Should an agent have its own identity?
- We are drowning in PDFs. Where do we start?
- Can we use our existing call recordings?
- Why has AI assistance not moved our cycle times?
- Why has AI coding tool adoption stalled here?
- What actually breaks after week two?
- Who is accountable when an agent gets it wrong?
Patterns we have shipped
Each of these is written from an engagement rather than from research, and each one states where our experience runs out.
Document intelligence to business intelligence
Getting information out of PDFs and into something the business can decide with.
You are here if the extraction demo impressed everyone and the business still will not act on the numbers.
Read the guide7 min readAI-native SDLC and product delivery lifecycle
Changing how software gets specified, built and shipped once AI is in the room.
You are here if the licences are bought, adoption has plateaued, and the delivery lifecycle has not changed.
Read the guide8 min readVoice agents and conversation intelligence
Turning conversations into structured intelligence, and holding conversations that take real actions.
You are here if the call recordings exist and nothing structured is coming out of them.
Read the guide6 min readEnd-to-end agentic workflow implementation
Taking one whole business process agentic, rather than assisting the humans doing it.
You are here if every step of a process now has an assistant and the process itself is unchanged.
Read the guide8 min readPatterns we would be approaching for the first time
We have not stood these up at enterprise scale. What we publish is the method, the failure modes we would design against, and what we are drawing on instead.
How it is built
The engineering underneath an agent: retrieval, tools, accuracy control, and choosing between them.
Retrieval, RAG and permission-aware knowledge access
Answering from your own knowledge, without answering from documents the person asking is not allowed to see.
You are here if the assistant answers well and nobody can prove it only answers from what the asker may see.
Read the guide12 min readMCP, tool calling and integrating agents with your systems
The moment an agent can call your systems, integration stops being plumbing and becomes an access decision.
You are here if the agent can now call your systems and nobody decided whose authority the call carries.
Read the guide14 min readGuardrails, hallucination and accuracy control
You will not stop a language model being wrong. You can decide in advance what happens when it is.
You are here if the answer to how you stop it being wrong is currently a paragraph in the system prompt.
Read the guide11 min readRAG, fine-tuning or prompting: how to choose
Sorting the requirement into knowledge, behaviour and cost, so the decision survives a finance review.
You are here if the architecture argument is being won by whoever read the most recent article.
Read the guide11 min readProving it and controlling it
Evidence, identity and governance. The parts that decide whether anything reaches production.
Agent evaluation and assurance
Why AI pilots never reach production, and what it takes to get one through the gate.
You are here if the pilot was judged by demonstration, and nobody will sign it off for production.
Read the guide10 min readAI governance and regulatory evidence
Building the evidence as a by-product of the work, rather than assembling it under deadline.
You are here if the evidence pack is being assembled by hand, after the fact, against a deadline.
Read the guide6 min readAgent identity and access
Agents are not users, and giving them a service account is how this goes wrong.
You are here if an agent is about to be given a service account and nobody can say what it is allowed to do.
Read the guide7 min readPaying for it
The money, the payback arithmetic, and what a finance function should refuse to accept.
If one of these is the thing you need bought rather than reasoned about, say so on the call and we will tell you whether we are the right firm for it.
Describe the problem and we will point you at the guide
Answers come from the 12 guides themselves, and the assistant will tell you first whether the pattern you are asking about is one we have delivered.
Prefer to talk it through? Ask us on a discovery call →
Running one of these?
Thirty minutes with James Rooney. We will tell you honestly which parts of it we have done before and which we would be doing for the first time.
Most organisations start with a fixed-price Agent-Readiness Audit · £30k–£90k · 6–8 weeks