The four modes of AI adoption
Most leaders treat AI as a switch: you have done it or you have not. It is not a switch. It is a ladder with four rungs, and almost every company is already standing on one of them. Here is what each rung looks like in practice, what it really costs, and where the work stops paying off.
If your team uses ChatGPT, your AI adoption has already started. That sounds obvious. It is also the single most misread fact in how companies think about AI, because it means the real question was never whether to adopt. The question is how far each part of your business can climb, and what each step up actually costs.
We sort that climb into four modes, ordered from the most hands-on to the most hands-off: Incremental, Assisted, Augmented, and Autonomous. They are not grades, and they are not a maturity badge. Each one is a real way of working that thousands of companies sit in right now. Higher is not automatically better. It is more autonomous, more expensive to stand up, and only worth reaching where the work justifies it.
A company is almost never on a single rung. Your support function might run at Mode 3 while finance is still at Mode 1. So the useful question is not "what mode are we", it is "which rung is each function on today, and which climb pays off next".
Mode 1: Incremental
A few AI tools, used by hand. Nothing connects. Someone pastes a customer email into ChatGPT, gets a draft back, copies it into the helpdesk, and sends. It helps. It also disappears the moment that person logs off, because the gain lives in one head and nowhere else.
The tools here are the familiar ones: ChatGPT, Microsoft Copilot, the assistant now baked into apps you already pay for. This is also where the "AI is basically free" idea lives. A single seat is about $20 a month, which feels like nothing, so everyone grabs one and the spend creeps up quietly with no one tracking what it bought.
Where it breaks: nothing compounds. No process gets faster, only individual people do, and only while they remember to. You are spending real money and you cannot point to what changed.
Mode 2: Assisted
AI drafts, a person finishes, and the output finally gets repeatable. The jump from Mode 1 is not better tools. It is that the whole team works AI-first and the first real workflows appear. Support replies start from a draft built on your own help docs. Sales follow-ups start from a shared template instead of a blank page.
Engineering feels this fastest, where tools like Claude Code or Codex turn a day of work into an afternoon. The cost is still mostly seats, plus the quieter cost of writing the prompts and templates the whole team leans on. That groundwork is cheap to start and easy to skip, and skipping it is the most common reason teams stall right here for a year.
Where it stalls: a person is still in every loop. Throughput is capped by how fast humans can review. To go further, the work has to run without someone triggering each step.
Mode 3: Augmented
Workflows run end to end. A person reviews instead of doing. The nightly run reconciles the day's orders, flags the three that look wrong, and leaves those for a human. Everything else is finished before anyone arrives. Your team stops doing the work and starts checking it.
This is the realistic ceiling for most companies inside a year, and it is the first mode where the bill is about more than seats. You now pay for orchestration, the tools like n8n that chain steps into a workflow, for the model calls each run makes, and for evaluations and monitoring so you can trust what happens while you sleep. The largest line item is usually the invisible one: the groundwork of clean data and written procedures that lets a workflow run safely. Skip it and the workflow does not merely underperform, it ships mistakes at machine speed.
The ceiling: every workflow is still something a person designed and watches. The system reacts well. It does not yet decide what to work on.
Mode 4: Autonomous
Agents run the routine and escalate the edge cases. People mostly sign off. The orchestration layer lives here: agents that pick up work, choose the next step, call the right tool, and surface only the exceptions a human needs to judge. The routine runs itself, and people spend their hours on the parts that genuinely need a person.
Mode 4 is really a range inside one rung. A lean version runs on off-the-shelf agents and a solid backbone. A bespoke version adds fine-tuned models, a dedicated operations layer, and the ability to swap the model under any task as prices and capabilities shift. That last point carries more weight than it sounds: the company that wins here is loyal to no single vendor. It plugs whatever is best this quarter into a foundation that does not move.
Where you stop, not where it breaks: autonomy is not a prize for its own sake. Some decisions should never run unattended, and a good plan names them out loud. The goal is not maximum automation. It is the right work running itself, with people kept on the calls that carry real risk.
The cost myth
The most expensive mistake in this whole picture is believing the low rungs are cheap. Microsoft 365 Copilot is a $30 per user add-on that sits on top of a business licence you are already paying for. ChatGPT Enterprise costs more again and will not sell you fewer than 150 seats. For a team of 40, that is $1,200 a month in Copilot add-ons alone, on top of the base licences, before a single workflow is connected.
That is not an argument against spending. It is an argument for knowing what you are buying. Cost rises as you climb, and so does the payoff, but the biggest expense near the top is not the software at all. It is the groundwork. That failure rate above is not a verdict on the models. It is what happens when companies buy at Mode 1 prices and expect Mode 3 results, with none of the data and process work that sits in between.
Where you actually stand
Here is what the ladder is really for. You are already on it, probably on several rungs at once across different functions, and you are almost certainly further along than it feels from the inside. The work is not deciding to start. It is seeing clearly where each function sits today, and which climb is worth making next.
We do not ask how ambitious you are, because that question flatters everyone and tells you nothing useful. We read where your functions already operate, from the tools you already run and the way your team already works, and we show you the next rung that earns its cost. Sometimes that is one step up. Sometimes a function is fine exactly where it is, and the honest answer is to leave it alone and spend the budget elsewhere.
A form can place you roughly in about ten minutes. The sharper, function-by-function read is what the assessment is for, and it is free to start.
Find out which mode each part of your business is ready for.
The free assessment places your functions on the ladder and shows the next rung worth climbing. About ten minutes, no card required. The full report is 450 EUR, and you only pay after you have seen the free version.
Start assessment