The Context Layer — an open standard, and the person who installs it
Your AIdoesn't knowyour business.
Nobody has written it down. Until someone does, every pound you spend on AI buys a very confident stranger.
The problem
You've probably already watched this happen
A system says something about your own company with total confidence, and it isn't true. A price you stopped offering in 2023. A process you abandoned after it went wrong. A commitment you'd never make.
Nobody spotted it, because it sounded exactly like something your company would say.
That's usually the moment an executive quietly decides AI isn't ready.
They're wrong about the cause, and the mistake is expensive. The model wasn't underpowered. It was uninformed. It has read most of the internet and none of your business.
Why the pilots failed
The gap isn't prompting, and it isn't tooling
You'll have seen the number — 95% of enterprise AI pilots deliver nothing measurable. I'd rather take it apart than lean on it, because you'd check. It comes from a July 2025 working paper out of MIT's Project NANDA: preliminary, not peer-reviewed, fifty-two interviews, a short measurement window, and written by a team who sell the infrastructure the paper concludes you need. Treat it as a mood, not a measurement.
The argument doesn't need it.
The companies building these models are among the most valuable ever created. And you probably cannot name one process in your own business that AI measurably improved this year. Neither of those needs a citation.
The technology is extraordinary. Almost nobody is making money with it. Both true, and the space between them is where your business currently sits.
It is not a mystery. It's a missing part.
Your company's real knowledge — how it prices, decides, escalates, refuses, and hands over — has never been written down anywhere a system can read. So there is nothing underneath the tools. Buying more capability doesn't fix an absence of substrate; it just makes the guessing faster.
The thing that's missing
Your business already has files and folders. They're stored in people.
How you price the awkward job that doesn't fit the sheet. Which enquiries are worth chasing and which are a polite no. What you'd tell a new starter in their first fortnight that isn't in any handbook. Why you walked away from that contract.
Every one of those is a file. Real, specific, valuable, used constantly — and held in one or two heads.
Which works, in the way that a business without a filing system works. Fine, until someone leaves, someone's ill, someone's on holiday, or you hire three people at once and discover the induction process is four people saying four different things in a corridor.
You already knew this. It's why holiday is stressful, why a resignation lands harder than the headcount justifies, and why "we should write that down" has been on the list for six years.
The AI question didn't create the problem. It attached a number to it.
What I build
The layer underneath
I write your business down in a structure your people and any AI system can both work from. Plain folders. Plain English. Markdown files anyone can read with no AI involved at all, and one entry file that tells any system where to look and what it's allowed to touch.
Then I build the first working systems on top of it.
The deliverable is disappointing to look at. That buys you four things.
- It's yours.
- No platform to rent from me. If you never speak to me again, you keep all of it and it still works.
- It survives the model.
- You'll change AI providers at least twice in the next three years. Everyone will. When you do, nothing underneath changes — you point the new one at the same structure and carry on.
- It can stay in the building.
- If your data can't leave the premises, the same layer runs against a model hosted locally. The structure doesn't care where the intelligence lives.
- It's readable by humans.
- So it's useful on day one, before any of the AI part, as the onboarding documentation you never got round to writing.
Convergence
I didn't invent this, which is the part I'd want to hear if I were you
Anthropic built Skills on this shape. Google's Open Knowledge Format does the same job over structured data. Every serious agent framework has arrived at the same idea from a different direction.
I'm not selling you a method. I'm installing the thing the industry has already agreed on, before your competitors get round to it.
And I've written the specification down and given it away. The Context Layer Standard is free, versioned, and licensed for anyone to implement or sell services against — including without me. If you have someone internal who can read a spec and do the work, take it and go.
Who's behind this
I'm Jack Horscraft, and there is no team
One person. Colchester, Essex.
I get a business out of its people's heads and into a structure a model can read. Pointed at marketing, that's an audience that finally converts — that's the work I do as The Second Layer. Pointed at the whole business, it's a company that runs without you. Same layer, different end of it.
I do this at a creative agency for eight-figure creator businesses, and I do it for SMEs directly.
This is a new product line, and I'm deliberately taking a small number of first implementations. I'd rather do three of these properly than ten badly, and the businesses that go first get more of my attention than anyone will get later.
What that buys you:
- My own layer, live. I run a consultancy, a day job, and client delivery off one of these. On a call I'll open it and answer a cold, hard question about a real client out of the folder structure, in front of you. That's the demo. There are no slides.
- The endpoints, proven and paid for. Content systems, conversion funnels, staff-training tools — built for real UK businesses on a version of this layer. They prove the endpoints. The layer as a whole product is newer than they are.
- The convergence argument above, which is why this doesn't depend on my judgment being right.
If you'd rather be one of the first, that's a conversation worth having now.
Where to start
Find out where you actually are
The Context Readiness Index is twelve questions and three minutes. You get a score out of 48, a band, and a two-page report naming your three weakest areas and what each one is costing you. No call, no obligation, and it's possible to score out of needing me.
It opens the week of 11 August. It's being calibrated against real businesses first, which is the only reason it isn't here now.
Read the Standard in the meantime, and implement it yourself if you've got someone who can. Or leave your email and I'll send you the Index the day it opens.
One email when the Index opens. Unsubscribe in one click, and I'd rather you did than sat on a list you don't read.
If you already know where you are, skip all of it and tell me what prompted you to look.
The actual question
Not "are we using AI yet"
Everyone is, whether you've sanctioned it or not.
The question is whether the thing your people are already using has ever been told how your business works.
Because if it hasn't, you don't have an AI problem. You have an extremely fast, extremely confident stranger with access to your customers.
And if the model you're using is replaced next year and you have to start again from nothing, you never had anything.
Build the layer underneath.