The Harness Map

title: "The Harness Map" dek: "Five stages of how a business actually uses AI, and the cliff between two and three where most companies stop short." date: "2026-03-25" pillar: "implementation" author: "floburn" tags: ["harness-map", "ai-implementation", "framework"]

Most businesses are behind on AI. They're stuck somewhere on the climb.

In the engagements we run, the first diagnostic question isn't do you use AI — that question has had the same answer for two years, yes, everyone does — but which stage of usage are you actually at? The answers vary widely and don't correlate with how loudly a company is talking about its AI roadmap. Some operating-stage companies don't talk about AI at all. Some companies that have been talking about AI for two years are still at stage one.

Here are the five stages of how a business actually uses AI, day-to-day. The framework is descriptive, not aspirational. It maps to what we see when we walk into a real operation, not what the corporate strategy deck says is happening.

1. Browser-tab AI

Someone opens ChatGPT, Claude, or a similar assistant in a browser tab when they need help. The use is personal, ad hoc, and invisible to the business. The employee gets a benefit; the company gets none. The benefit is real to the individual — drafting an email, summarizing a doc, brainstorming a presentation — but it doesn't accrue to the organization in any structured way.

The risk at this stage is symmetric to the benefit. Shadow tooling means data the company can't track is leaving the building, going into models the legal team hasn't reviewed, on accounts the IT team doesn't see. The leverage is invisible. So is the exposure.

Most companies are here whether they know it or not. Many are surprised, on audit, how deep into Stage 1 their employees already are.

2. Tool-assisted AI

A handful of point tools. A writing assistant. A meeting summarizer. Maybe a Zapier hookup that does something useful in one workflow. Saves time in pockets. Doesn't change how the business runs.

This is where most companies stop after their first AI investment cycle. Two SaaS subscriptions, three named user accounts, one quarterly slide deck about our AI strategy. The savings are real and measurable in hours per week per user, but they don't compound. Each tool is bounded by its own subscription and its own users. The business operating model is unchanged.

Stage 2 is also where almost every consumer-facing AI product wants you to live forever. The unit economics of a SaaS-per-task pricing model depend on you being here, not climbing higher.

3. Workflow AI

AI is wired into one or two real workflows — intake, research, customer responses, scheduling, content production. The work itself looks different now. Cycle time drops. Throughput rises. Variance falls. The people who used to do that work either do something else or do that work at a different scale.

Most businesses that say they're doing AI stop here. Some never get here at all. The gap between Stage 2 and Stage 3 is the cliff.

The reason the cliff exists is structural. Stage 2 work is buying tools. Stage 3 work is redesigning a workflow, which is a different category of organizational labor. It requires someone with authority over the workflow to commit to the redesign, a build phase that produces something more than a configuration change, and a measurement framework that didn't exist before the AI was inserted. The tools to do Stage 3 have existed for two years. The internal capacity to use them is what most companies lack.

4. Agent AI

Specific, named agents do specific, named jobs — and they do them on a schedule, on a trigger, or in response to events the business generates. The lead-research agent. The support-triage agent. The compliance-attestation agent. Built once, run forever, owned by the business, monitored like a junior employee.

Stage 4 is rare. Most companies that think they're at Stage 4 are actually at Stage 3 with an internal naming convention. The test is operational: does the agent run when no one is watching? Does its output go into systems of record? Does someone get paged when it fails?

When Stage 4 is real, the company experiences a step-change in capacity. A two-person ops team starts producing what would have required four. A small support team handles a customer base sized for a much larger team. The leverage is structural, not personal.

5. Operating AI

AI is part of how the business runs. New employees are trained on it. New processes are designed around it. Pulling it out would be like pulling out email.

Stage 5 isn't bigger than Stage 4. It's deeper. The business has internalized AI as a load-bearing component of its operations. Process redesign happens with AI as an assumed capability, not an addition. New hires don't learn how to do their job without AI and then have AI grafted on; they learn the job as it's now done.

Few companies are here. Those that are tend not to advertise it, because the advantage is operational, not narrative. The risk profile shifts, too. The dominant risk at Stage 5 is concentration — model degradation, vendor changes, cost shifts. A Stage 5 company has to plan for the absence of the model the same way a coastal city plans for storms.

The cliff

Where most people get stuck: the gap between 2 and 3. The tools exist. The capability exists. What's missing is someone whose full-time job is to close that gap — which almost nobody on the inside has time to be.

The companies that successfully cross the cliff usually do one of two things. Either they make crossing the cliff someone's actual job — pulling a senior operator off other duties and giving them a quarter to redesign one workflow end-to-end — or they bring in outside help whose only assignment is the crossing. The work doesn't happen in the margins of someone's existing role. It can't. The cliff is structural; it requires structural commitment to cross.

What it doesn't require is more tools. The companies that try to climb the cliff by buying more Stage 2 tools find themselves at Stage 2.5, with a higher monthly bill and the same operating model. The output of more Stage 2 spending is more Stage 2 leverage, which is to say more time savings for individuals, distributed thinly. The output of Stage 3 work is a different business.

What this map is for

The Harness Map is a tool we use in the first 30 minutes of every diagnostic engagement. It's not aspirational. It's a way to make sure the conversation starts with shared vocabulary about where the business actually is, not where the strategy deck says it should be.

When we walk into a room and use the map, three things tend to happen. First, the leadership team disagrees with each other about which stage the business is at — and the disagreement is the data. Second, the rooms that haven't done this exercise underestimate how much of the company sits at Stage 1. Third, the rooms that have done some version of this exercise overestimate where Stage 3 begins.

If you want to do this exercise on your own, the Readiness assessment is a self-administered version — 12 prompts across the five stages, no email capture, no follow-up sequence. Run it before the conversation. The map gets sharper when both sides have answered honestly first.

If the diagnosis points at Stage 2-to-3 — which is the conversation we have most often — that's the work we do. A discovery call is the right first step. We won't recommend the work if it doesn't fit. We will tell you what stage we think you're at, and what would close the gap.