Layer 1: Context, or How to Structure a Business So AI Can Actually Help Run It

Every AI install that stalls in month four has the same root cause: Layer 1 was never real. The automations are wired. The dashboards render. The intelligence brief ships on schedule. None of it knows the business it is operating inside, because the business was never written down in a form the system could read.

Context is Layer 1 of the AIOS. Every later layer is parameterized by it. Data pulls against the ICP definition. Intelligence summarizes against the tier definitions. Automate applies the decision-rights matrix to approval gates. Build decides what is custom and what is standard by reading the nonnegotiables. If Context is vague, the downstream layers are guessing, and guessing scales badly.

This post is about the technical shape of Context. What it is as a file. What goes in it. What stays out. How we capture it. How we know it is done.

What Context actually is

Context is not a slide. It is a document.

Specifically, it is a structured markdown artifact, versioned in a repository the leadership team can read and edit. No proprietary format. Markdown works because it is readable to humans, parseable to machines, diffable in version control, and portable across whatever tool the team uses today and whatever replaces it in two years.

Three properties matter more than the format.

Versioned. Every change is a commit with a timestamp and an author. You can ask "what did our ICP look like six months ago" and get an exact answer.

Structured. Sections are consistent across clients. Not because we force the same strategy on every firm, but because a system downstream needs to know where to find the ICP definition, the decision-rights matrix, and the voice rules without parsing free-form prose.

Living. The document is updated in Run. New hires read it on day one. Quarterly leadership sessions revise it. When a policy shifts or a tier gets added, the change lands in the document first, and only then in the systems downstream. If Context drifts out of sync with reality, every layer above it drifts with it.

A typical mid-market Context artifact runs fifteen to thirty pages. Longer is a sign the team is describing aspiration. Shorter is a sign the interview was not thorough.

What goes in the Context document

The sections are stable across engagements, and each one is scoped to answer a specific operational question the downstream layers will ask.

Strategy statement. One or two paragraphs. What the business does, who it does it for, and why it wins. Not a vision statement. The version leadership could say out loud on a sales call without notes. If three leaders write it separately and produce three different versions, it is not ready.

ICP definition. Specific enough to act on. Industry, company size band, buyer role, the triggering event that makes them start shopping, and the disqualifying signals. "Mid-market operators" is not an ICP. "Non-regulated professional services firms, $3M to $25M revenue, founder or second-generation operator, triggered by a key senior hire leaving" is. Data pulls against this definition. Intelligence scores against it. If it is loose, everything downstream is loose.

Commercial model. Pricing bands, engagement structures, renewal mechanics, referral economics, margin targets by line of business. Enough detail that a new analyst could correctly price a typical deal without asking.

Voice and tone rules. Drafts that go out under a partner's name have to match the partner's voice, and the rules have to be concrete enough to enforce. Not "friendly and professional." Sentence length bands. Banned phrases. Required framing. The specific shape of how the firm opens and closes client emails.

Client tier definitions. If the firm treats a $500k-ARR client differently from a $50k-ARR client (it does), those differences are written down. Response-time commitments, senior involvement, meeting cadence, custom work eligibility. The intelligence layer uses tier definitions to decide what to surface for whom.

Decision-rights matrix. Who owns which call. Pricing exceptions, scope changes, hiring, firing a client, public positioning, new service lines. The automate layer uses this to build approval gates: a request routes to the person the matrix names, not to whoever happens to be online.

Hard nonnegotiables. What the firm does not do. Client types we decline. Work we refuse. Positioning we reject. The most important section for the build layer, because it is the only reliable way to tell the system "do not propose this, even if it looks like a win."

What stays out

The discipline of Context is as much about exclusion as inclusion.

Anything that changes weekly. Pipeline numbers, utilization, cash position, project status. These live in Layer 2. If it moves faster than quarterly, it does not belong in the Context document. Context describes the shape of the business. Data describes its current state. Mixing them produces a document that is out of date the week it is finished.

Anything aspirational rather than operational. "We want to move upmarket" is not Context. "We sell to firms $3M to $25M in revenue" is. When the firm actually moves upmarket, the ICP changes and the document is updated. Until then, the document describes the business that exists.

Anything no human would enforce. A voice rule that nobody will push back on in review is not a rule, it is wishful thinking. A nonnegotiable that a partner will override on the third exception is not a nonnegotiable. Before a line goes in, we ask whether a leader would actually block work that violated it. If not, it stays out.

How we capture it

Context does not get written by a consultant in a room. It gets captured from leadership and drafted back to them.

The Context interview is three to five focused sessions with the leadership team over two to three weeks. Each session is scoped to a specific section. Strategy and ICP in the first. Commercial model and tier definitions in the second. Voice, decision-rights, and nonnegotiables in the third. One or two sessions to reconcile contradictions the first three surfaced.

We ask questions designed to pull out what already lives in people's heads. "Walk me through the last deal you closed and the last deal you disqualified." "Who decided on the last pricing exception, and what did the approval conversation look like?" "Read me the last three client emails you sent, and tell me what you edited out of the drafts." The job is to surface operating reality, not to generate strategy.

Between sessions, we draft. Leadership red-pens. We reconcile. The whole capture runs two to three weeks for a typical mid-market firm, and it is the single most leverage-dense piece of work in the Blueprint phase.

The principle the AAA methodology keeps returning to applies hardest here: start with context, not tools. The capture session is deliberately vendor-free. Nobody is deciding which platform to use. We are writing down the business.

The legibility test

Context is done when the document passes the legibility test.

Hand it to a new hire on their first day or to a new AI assistant. Ask them the three to five most common operating questions the firm fields every week. "Which leads should we prioritize?" "Does this request need partner approval?" "What tier is this client and what service level do they get?" "Is this engagement in scope for us or out?"

If the document produces correct answers without them asking a human, Context is real. If they have to escalate more than once, the document is not done. The gap tells you exactly which section is underspecified.

This is not an abstract test. Before we sign off on the Context layer, we run it. Sometimes with an actual new hire. Often with an AI assistant pointed at the document and asked the firm's ten most common questions. The acceptance bar is concrete and measurable. We wrote more about how we set acceptance bars for each layer in the acceptance bar for a layer.

Why skipping Context kills pilots

The most common failure pattern is a leadership team that skips or shortchanges Context to save time, usually because a vendor has convinced them that the automation layer is where the wins live.

The pilot looks fast for ninety days. Automations fire. Dashboards light up. Then month four arrives, and the system starts producing outputs the team cannot use. Lead scoring prioritizes the wrong accounts because the ICP was never codified. Intelligence briefs summarize against tier definitions that do not exist. Approval gates route to the wrong partners because the decision-rights matrix was never written down. The team starts overriding the system more often than it trusts it, and the install degrades into expensive shelfware.

We described the mechanics in why AI pilots die in month 4 and the decision-pattern angle in AI readiness is about decision patterns. The short version is the same every time: the AI was installed on top of a business it did not know. The business was in the CEO's head, in tribal knowledge, in Slack threads nobody could find. The system cannot read any of that.

The other version of the same failure is the tools-first version: teams route around missing Context by piling more software on top. It does not work. We covered why in why more software makes operations worse and the real cost of spreadsheets and Slack. Tools are downstream of Context. Skipping Context to install tools is a category mistake.

The AAA principle is simple. Install in layers, not in leaps. Each layer earns the next. Skipping Layer 1 does not save time. It moves the cost from the front of the install to the middle, where the cost is higher and the credibility is thinner. HBR's work on operations strategy and MIT Sloan's AI research keep landing on the same finding: the companies getting durable operating leverage from AI are the ones that wrote the operating model down first and wired the AI to it.

Layer 1 is also what the sibling posts in this series point back to. Layer 2 covers data centralization. Layer 3 covers the morning brief. Layer 4 covers approval gates. Layer 5 covers custom vs. module calls. The emerging voice-first operations work depends end-to-end on a real voice-and-tone section here. None of those layers holds without Context underneath.

The Fit Check tells you whether the Blueprint is worth running. The Blueprint is where the Context document gets drafted. The Build phase is where the downstream layers get installed against it.

Start at Layer 1. Every hour you spend there buys weeks of runway above it.

-Jeremy