Service 04 · The AI Harness

The AI Harness — your AI, your data, your rules.

One clever agent is a party trick. A harness is a business. It is the layer that connects your models, your data and your rules into a single system that is orchestrated, monitored and maintained — so the company runs on AI rather than experimenting with it.

First, A Clarification

Not the developer’s harness.

If you have seen the phrase “AI harness” in engineering circles lately, it usually means the scaffolding a programmer wraps around a coding assistant. That is a real and useful thing, and it is not what this page is about.

Here, an AI Harness means something at the business level: the connective layer that turns scattered AI experiments into one system your company can actually depend on. Same word, different altitude — worth saying plainly so nobody arrives expecting the other one.

The Problem It Solves

Everyone adopted AI. Almost nobody connected it.

The common picture is not an absence of AI — it is a scattering of it. Someone in sales has a subscription. Someone in operations built a clever automation nobody else knows about. There was a pilot last year that impressed everybody and then quietly stopped. Three different tools each hold a piece of the same customer.

Every one of those is genuinely useful, and together they still do not add up to a business that runs on AI. Nothing shares context. Nothing shares rules. When something goes wrong, nobody owns it. And the value stays stubbornly personal — it belongs to the individual who set it up, and it leaves when they do.

A harness is the layer that makes them one thing. Shared data, shared rules, one place where it is watched, and a person accountable for it. We wrote about this gap at length in Cross the Gap.

What Is In It

The five layers.

Your models

Whichever engines suit each job, behind one interface, swappable as the field moves — and it moves constantly.

Your data

The records, documents and history the AI must reason over, connected properly rather than pasted into a prompt.

Your rules

What it may decide alone, what needs a person, what it must never do. Written down once and enforced everywhere.

Orchestration

Several specialised agents with something sensible above them, routing work and keeping context between steps.

Watch & recover

Monitoring that proves it is alive with a real call rather than assuming, alerts a human when it is not, and fails over instead of going quiet.

An owner

A named person accountable for the whole thing — the layer most AI programmes are missing entirely.

We Run One

This site is on the end of ours.

Ray is not a bought widget. He runs on our own harness: model orchestration, a live connection to the business data behind this site, rules about what he may say and what needs a human, an audit trail of every conversation, and monitoring that proves the connection is alive by actually using it rather than trusting a clock.

When part of it fails, it fails over to a spare credential and tells us. That is not a slide in a deck — it is the thing serving you this page. The same architecture is what gets built for clients, sized to their problem. See what it has been used for.

How You Take It On

Start small. Start properly.

1

Nobody starts with a full harness. It grows out of the first thing that works — one automation or one agent that earns its place, then connects to the next.

2

Host it yourself or have us run it. Both are real options with real prices, not a nudge towards the one that suits us. See plans.

3

It is watched, not just delivered. A harness nobody is monitoring is a harness that is quietly broken and will be discovered by a customer.

4

You own all of it. Code, data, keys, documentation. If you want to take it in-house, that is a handover rather than a negotiation.

Questions

The ones we actually get asked.

What is an AI Harness?
At Strange Materials it means the connective layer that turns scattered AI tools into one system a business can depend on: your models, your data and your rules, orchestrated together, monitored, and owned by a named person. It is distinct from the developer-tooling sense of the phrase, which refers to scaffolding around a coding assistant.
Do we need one, or is a single automation enough?
Start with the single automation. A harness is what you grow into once two or three AI things exist and start needing to share data, rules and supervision. Building the connective layer before there is anything to connect is a way of spending money on architecture nobody uses yet.
Does a harness lock us into one AI provider?
The opposite — that is one of the main reasons to have one. Models sit behind a common interface so they can be swapped as capability and pricing change, which they do every few months. Anything welded to a single provider is a liability by design.
Who looks after it once it is live?
Either your team or ours, and that is a decision you make rather than a default. Support levels and the option to host it yourself are set out on the plans page. Whichever you choose, monitoring and the named owner are part of it, because an unwatched system is an outage waiting to be reported by a customer.
Can it work with the AI tools we already pay for?
Usually, yes, and that is normally the cheapest route. The harness is a connective layer rather than a replacement, so existing subscriptions and automations that are genuinely working get connected in rather than thrown away.
The Rest of the Lab

These combine.

Most builds use two or three of them together. See all services.

Step One

Start with one honest sentence.

“Here’s what’s quietly eating my week.” Tell Ray in your own words. A clear, priced build path comes back, and a human signs off everything that leaves the lab.

Start the scan — free
Free scan · Fixed prices · You own everything