Five layers that make AI stick: graph, loop, harness, context, prompt

Most AI projects start at the bottom. Someone finds a good prompt, connects a tool, adds some documents, and later wonders why it never became part of daily work.
We go the other way. We work through five layers from the top down, then loop back to the top. This is the method behind everything we do. It is why we call the result AI-native, not AI-assisted.
Key points in 30 seconds
- Order matters. Map the work, measure it, add guardrails, add your knowledge, and only then write the prompt.
- The prompt goes last. It is the cheapest layer to change and the least likely to be the real problem.
- Each layer makes the next one cheap. If you skip one, the layers below must make up for it, badly.
- Results feed back. What each cycle shows updates the map, so the process improves without a new project.
- Design top-down, fix bottom-up. Both are true at once.
What are the five layers?
Tap a layer to see the question it answers, what goes wrong without it, and what you get.
Graph engineering
1. Map the work
How does work really flow? Which steps, decisions, hand-offs and owners?
Without it
You automate the wrong thing. A tool fixes one task, then the next step breaks, or work piles up at a person who was never told.
What you get
A list of steps, who owns each, where decisions sit and where work waits. It also shows steps that should not exist. Removing those is often the cheapest win.
Loop engineering
2. Close the loops
How will we know a change helped? Who reads the result every cycle?
Without it
You cannot tell if it worked, however good the demo looked.
What you get
One number per process (for example cycle time, error rate or cost per case) and one named person who reads it every week. If the number does not move, the change did not help.
Harness engineering
3. Build the harness
What can the agent do, see and spend? How do we stop it?
Without it
It is not safe to switch on.
What you get
Everything around the AI model: the tools it can call, the data it can read and write, its permissions, a spend cap, an off switch, a log, and a person who signs off before anything reaches a customer.
Context engineering
4. Feed it your knowledge
What must the agent know about this business to do the step well?
Without it
The AI knows the internet, not your business.
What you get
Your price list, your policies, how you handled the last ten hard cases, and the exceptions everyone knows but nobody wrote down. The work is to find it, clean it and organise it so the agent finds the right piece at the right step.
Prompt engineering
5. Tune the instructions
How is the agent told to do this one step?
Without it
A clever prompt cannot rescue a project that is missing the four layers above it.
What you get
With a mapped step, a measure, a harness and the right context, the instructions are usually short and easy to get right. This is the easiest layer to change, so it goes last.
The most common failure we see is a tool added to one task. It works. Then the next step breaks, because nobody checked what that step expected to receive.
Should I design from the top or the bottom?
It depends on the job. Our piece on harness, loop and graph engineering suggests the opposite order. That is not a contradiction. Switch the view below.
Start at the top. Know the work first.
- 1Map
- 2Measure
- 3Guardrails
- 4Knowledge
- 5Instructions
You must know what the work is before you decide what any agent should do.
What is the loop back to the top?
Every cycle, the measures show something. An exception an agent handled badly. A step that still waits on a person. A hand-off nobody needs now.
That finding updates the map. The change then flows down through the measures, guardrails, knowledge and instructions.
People call this recursive self-improvement. It sounds bigger than it is. In practice, it is a weekly meeting with a number on the screen and the power to change the map. Nobody has to start a new project. Your own team runs the loop.
What happens in the first 90 days?
We join your standups and decisions, map how the work really flows, and write a plan and a cost the owner can defend.
Measures and feedback on the chosen process, agents with guardrails, the knowledge they need, and their instructions. The first AI-run process goes live, measured and capped.
Your team is trained as we build, and takes over the weekly loop. The map, the agents, the playbook and the data are already in your accounts.
Working with us
If the change is in how your business runs, this is what a fractional CAIO does with us. If it is in a product you build, it is the same method, led by a Forward Deployed CTO. The Method page shows what each layer leaves you with.
Not ready to talk? Get your free AI operating plan in two minutes. It is a first map of where AI fits in your business, and it runs in your browser.
Frequently asked questions
What does it mean for a business to be AI-native?
It means the business is designed so that AI agents do the repeated work and people make the decisions, with a measure on every process and a loop that improves it. It is not the same as everyone having an AI tool. A company can buy licences for every employee and still run exactly as it did before; an AI-native company has changed how the work flows.
Why start with graph engineering instead of prompts?
Because the prompt is the cheapest layer to change and the least likely to be the real problem. If nobody has mapped how the work flows, the agent is automating a step that may not need to exist, handing off to a step nobody owns, or waiting on a decision nobody wrote down. Mapping the work first tells you where AI belongs at all. Prompts are tuned last, once the layers above them exist.
What is the difference between context engineering and prompt engineering?
Context engineering decides what the AI knows when it does a piece of work: which documents, data, past decisions and memory it can reach, and in what shape. Prompt engineering decides how it is instructed to use that knowledge for one step. Good context with an average prompt usually beats a brilliant prompt with no context, which is why context comes first.
What does recursive self-improvement mean for a small business?
Nothing to do with runaway AI. It means the process improves itself in a plain, measurable way: every process has a measure, every cycle someone reads it, and what it shows updates the map, the agents and their instructions. After a few cycles the business runs better than it did when the work was handed over, without anyone starting a new project.
Do we need a software team to do this?
No. The first two layers — mapping the work and closing the loops — need no code at all. The harness, context and instructions can be built with off-the-shelf agent platforms for most operations work. You need someone who owns the result and understands how the layers fit together, which is the role a fractional Chief AI Officer plays.