AI Literate, AI Enabled, AI First: The Adoption Roadmap That Actually Sticks

An executive at a firm with a few thousand staff asked us last month: "How do we roll AI out to everyone without leaving half the company behind?" Then, before I could answer, the real question came out: "And are we already behind?"

Those two questions travel together, and the honest answer to the second one makes the first one easier. Most companies are further back than they think everyone else is. The gap between the AI headlines and the median business is enormous, which means the race is still winnable. But only if you stop treating adoption as a software rollout and start treating it as a maturity path with stages you can actually check.

People call this AI transformation. I find that framing makes it sound like one big leap, when in practice it is three distinct stages: AI Literate, AI Enabled, AI First. Each stage has concrete markers. If you cannot point at the marker, you are not at the stage, no matter what the board deck says.

What Are the Three Stages of AI Adoption Maturity?

AI Literate means everyone can use AI safely and most people actually do. AI Enabled means AI is embedded in named workflows with owners and measured outcomes. AI First means new processes are designed for AI agents from the start, with humans specifying and reviewing the work.

The stages are sequential for a reason. Companies that try to jump straight to agent-driven workflows on top of a workforce that has never been trained end up with a handful of power users, a lot of quiet resentment, and a security incident waiting to happen. Companies that stall at literacy get a nicer intranet and no P&L impact. The value compounds when each stage builds on the one before it.

Here is what each stage looks like when it is real.

What Does AI Literate Actually Look Like?

AI Literate means three things are in place: everyone has access to an approved AI tool, everyone has had basic training in their own role's context, and a safe-use policy exists that people have actually read. The marker that tells you the stage is working: sustained weekly use across roles, rather than licences sitting idle.

That measure matters because it separates literacy from theatre. Plenty of organisations have bought licences and run a lunch-and-learn without building a repeatable habit. If weekly active use remains concentrated among a small group, the rollout has not yet reached the wider organisation.

Two things move that number. The first is training that uses the person's real work, not generic prompting tips. A property manager should leave their first session having drafted an actual arrears letter. The second is a safe-use policy that says yes more than it says no. If the policy is a wall of prohibitions, people will use their personal accounts on their personal phones and you lose all visibility. Google reported Gemini crossing 900 million monthly users this year; your staff are already using AI at home, so the question is not whether they use it, it is whether they use it inside your guardrails or outside them. I have written more on getting that balance right in freedom inside guardrails, because the policy is what makes the access safe to give.

What Does AI Enabled Mean, and How Is It Different from Literacy?

AI Enabled means AI is embedded in named workflows, each with an owner and a measured outcome. Not "the team uses ChatGPT sometimes". Rather: "Monthly client reporting is AI-drafted and human-reviewed, Sarah owns it, and turnaround dropped from four days to one."

The unit of progress at this stage is the workflow, and the workflows worth targeting are rarely the ones people expect. Anthropic's own usage data from Claude Cowork, drawn from 1.2 million anonymised sessions across 600,000 organisations, shows business-process operations (reporting, checklists, spreadsheet reconciliation) at 33.4 percent of usage, while software development sits at 8.7 percent (source). Read that again if your AI programme currently lives inside the IT department. Adoption is now led by general knowledge work. The transformation belongs to the finance team, the operations team and the client services team before it belongs to the engineers.

Getting to AI Enabled requires a champions network, and this is where most rollouts die, so I want to be precise about why. Champions fail when they are volunteers doing AI advocacy on top of a full day job. That is not a motivation problem, it is a resourcing decision that leadership dodged. If a champion has four hours a week of protected, rostered time to build and document workflows for their team, you get workflows. If they have zero protected hours and a Slack channel, you get a Slack channel.

The marker for AI Enabled: you can list your AI-embedded workflows by name, name the owner of each, and show the before-and-after numbers.

What Does AI First Look Like in Practice?

AI First means that when a new process is designed, the default question is "which parts of this does an agent do?" rather than "which parts could we maybe automate later?" Humans specify the work, review the output, and handle the judgement calls; agents do the production. The marker: work across the business is shaped for delegation by default, with briefs, checklists and review gates built in from day one.

The tooling is now built for exactly this pattern. Claude Cowork went cross-device in July, from desktop-only to web and mobile (source), which makes "kick off a task, monitor it, approve the result" the default shape of a working session rather than a novelty. The bottleneck in an AI First company moves from doing the work to specifying and reviewing it, and that is a skill you build at stage two, which is why you cannot skip it.

AI First also depends on something less glamorous: your company's knowledge being somewhere an agent can use it. Agents are only as good as the context you give them, and most businesses keep their context in the heads of four senior people. That is a solvable problem, and I have laid out the approach in building the company brain.

How Do You Move a Large Organisation Through This Without Leaving Anyone Behind?

Set a uniform floor and let the ceilings vary. Every person in the business, from the receivables clerk to the CFO, gets to the literacy floor: access, training in their context, weekly use. Then each function climbs towards AI Enabled at its own pace, guided by its own champion, on workflows it actually owns.

Trying to move everyone at the same speed on everything is how you leave people behind, because the programme calibrates to the enthusiasts and everyone else quietly checks out. What works instead:

• Track sustained weekly use per team, with managers owning the measure rather than relying on a company average that hides departments where adoption has stalled.
• Run training in cohorts by role, using that role's real documents and real tasks. Accountants train on reconciliations, not haiku.
• Pair every reluctant team with a champion from a similar team who has a working result to show, because peers persuade where mandates do not.
• Publish the workflow wins internally, with the owner's name on them. Recognition is cheap and it compounds.

The people most at risk of being left behind are rarely resisting the technology. They are usually the busiest, most operationally loaded people in the building, which brings us to the resourcing question.

How Do You Stand Up the Internal Motion When Everyone Has a Day Job?

You buy the time. There is no version of this where a serious adoption motion runs on discretionary effort, so the real decision is what you buy: internal headcount, protected time for existing staff, an external partner, or a mix.

Protected champion time is the non-negotiable baseline, and it is cheaper than it looks. Ten percent of one person's week per team, formally rostered, survives busy periods; "when you get a chance" does not.

On the hire-versus-partner question, I will be straight with you even though Ential is an external partner. Hiring an internal AI lead is the right call when you are past 100 staff, you have workflows live and producing, and the role has a full-time pipeline of work rather than a mandate to "drive AI". An internal hire compounds context about your business that no outsider ever fully matches. The failure mode is hiring that person first, before any workflows exist, and asking them to generate momentum alone against 15 department heads who did not ask for them.

A partner earns its keep in the earlier stages: standing up the literacy programme, finding the first high-value workflows, training the champions, and building the review habits, because a partner has run the same motion in a dozen businesses and knows which failure is coming next. The honest pattern I recommend to most mid-market companies: partner to build the motion, hire to run it, and make the partner's exit criteria explicit from the start. Anyone selling you a permanent dependency is selling you something other than adoption.

How Do You Get People Excited Instead of Scared for Their Jobs?

Show each person their worst task disappearing first, and be honest about job redesign instead of hiding behind slogans. "AI won't replace you, someone using AI will" is a threat wearing a slogan's clothes, and your staff hear it that way.

What actually shifts the room is specificity. Sit with a team, ask which task they would pay money to never do again, and automate that one first. When the month-end reconciliation that ate someone's weekend now takes 40 minutes of review, the fear conversation changes on its own, because the person has experienced AI taking the part of the job they hated and leaving the part where their judgement matters.

Then say the honest thing out loud: roles will be redesigned. Some tasks are going away permanently, the role that remains is more review, more judgement, more client contact, and here is what we are committing to do with the time that gets freed. People can handle a straight answer about change. What they cannot handle is a leadership team that clearly knows change is coming and keeps insisting nothing will.

Are We Behind? What Is Everyone Else Actually Doing?

Almost certainly less than you fear. Most companies, including impressive-sounding ones, are at stage one: licences bought, usage patchy, no named workflows, no measured outcomes. The polished case studies you read are the outliers, not the median.

Here is the more useful framing. The frontier moved again just last month, new model tiers, new agent capabilities, new pricing, and it will move again next quarter. In an environment that resets that often, the compounding advantage does not go to whoever bought tools fastest. It goes to whoever built the learning motion: the trained workforce, the champions with protected time, the habit of turning one team's workflow win into every team's workflow. Tools depreciate in months. The motion appreciates for years.

So no, you are probably not behind. But the cost of staying at stage one is now growing every quarter, because the companies that built the motion are compounding while everyone else is still running pilots.

Where to Start This Quarter

Pick the stage you are honestly at, then chase that stage's marker and nothing else. If weekly use is not sustained across roles, you are at stage one and your job is access, training and a safe-use policy. If literacy is real, your job is three named workflows with owners and numbers. Everything else is noise.

If you want help building the roadmap for your business, that is exactly what our AI consulting engagements do: we assess where you actually are, design the motion with your people, and build it so it keeps running after we leave. No pressure either way; the model above works whether or not we are in the room.