You were given a Ferrari. Most people are still using it as a shopping trolley.

Imagine someone handed you a shopping trolley and you figured out pretty quickly that it was useful for carrying things around, so that is what you used it for. Meanwhile, the people who built the trolley kept quietly adding to it, an engine here, better steering there, a whole new set of capabilities tucked in underneath, but because it still looked like a shopping trolley and you were still putting groceries in it, you never really stopped to ask whether it might do something more. That is more or less what has happened with AI over the last two years, and most businesses are still pushing their trolley around the car park wondering why everyone else seems to be getting somewhere faster.

When ChatGPT launched in late 2022, it was genuinely remarkable for what it was: a tool you could have a real conversation with, that knew things and could explain them clearly, that felt less like a search engine and more like a very well-read colleague. So people started using it the way you would use a well-read colleague, asking it questions, getting answers, copy-pasting the useful bits into whatever they were working on. A Google that you can have a conversation with, as someone put it to me recently, which is a fair description of what it was at the time.

The problem is that what it was at the time and what it is now are genuinely different things, even though nothing about the interface has changed. The chat window looks the same. The experience of typing a question and reading a response feels the same. But the capability sitting behind that familiar interface has been rebuilt, extended and quietly transformed to the point where describing the 2022 version and the current version as the same product is a bit like saying a bicycle and a Formula One car are the same because they both have wheels and get you from A to B.

What the original version was actually good for

In its early form, AI like ChatGPT was genuinely useful for a specific category of task: generating text. Draft me an email, summarise this document, explain this concept in plain English, give me five ways to say this thing differently. These are real time-savers and they are still worth doing, but they all share the same underlying structure. You give the AI an input, it gives you an output, and then a human takes that output and does something with it. The AI produces, the human acts, and there is always a person in the middle carrying the result from one step to the next. That is the shopping trolley. Useful, worth having, but it needs someone to push it.

What it became while you were not paying attention

Over the last two years, while most people were still asking ChatGPT to rewrite their emails, the underlying tools were acquiring a set of capabilities that changed their nature entirely. They can now browse the internet, read and interpret files, write and execute code, connect directly to other software systems, and take real actions inside applications. More significantly, they can chain all of these capabilities together into sequences. Not just "do this task" but "do this task, and then based on what you find, do this next thing, and if this condition is met, trigger this other step, and when it is done, send the result here." That is a workflow, not a conversation, and it is a fundamentally different relationship between the tool and the work.

This is where the Ferrari has been sitting, in the same browser tab, the whole time.

What the difference looks like when it matters

A business owner still using AI as a search engine might ask it what a good contractor onboarding process looks like and then go and build that process themselves. A business owner who understands what the tool has become might instead build a workflow that receives a new contractor application, checks the submitted documents against compliance requirements, follows up automatically on anything missing, updates the internal system once everything is in order, and notifies the relevant manager when the contractor is ready to go, with no human involved until something genuinely needs a judgment call. Same tool, completely different relationship with it, and the difference in outcome is not marginal. One saves ten minutes of research. The other eliminates a process that was previously eating half a day of someone's time every time it ran.

We have built exactly this kind of workflow for clients across a range of industries. A building compliance firm that was spending 90 minutes per job on manual data entry now processes the same job in three minutes, with AI reading the building plans, populating the forms, and flagging anything that needs a human eye. A vocational college that was manually connecting enrolment forms to their student management system scaled from 200 to 2,000 students without adding a single admin role. The tool doing the work in both cases is not fundamentally different from what is sitting in your browser tab right now. What is different is how it is being used. If you want to understand how we map these opportunities, our FlowTrace process is a good place to start.

Why most people are still at step one

It is not a matter of intelligence or technical ability. It is that nothing ever forced people to update their mental model of what the tool had become. When software updates with new features, you usually know about it. There is a notification, a tutorial, a what's new screen that walks you through what changed. AI development does not work like that. The new capabilities arrived gradually, folded quietly into the same interface people were already comfortable with, and there was no moment where the tool announced that it had crossed into genuinely new territory. So people kept using it the way they always had, which worked fine, but which left an enormous amount of what the tool had become sitting unused.

What this means for your business right now

Almost every business has processes where a person is currently moving information from one place to another, applying a routine rule, and passing the result along to the next step. If you want to understand what that is costing you, the maths are worth doing. These are exactly the kinds of processes that the current generation of AI is designed to handle, and the gap between "we use AI to help with a few tasks" and "AI is running parts of our operation" is really just the gap between knowing what the tool used to be and understanding what it has become. The shopping trolley is still there if you want it. But there is an engine in it now, and most of your competitors have not worked that out yet either.