The Different Stages of AI Usage

As AI becomes commonplace, a pattern is emerging: some people don't use it at all, some use it regularly, and a few seem to bend it entirely to their will. With that in mind, I think there are four stages a person moves through on their AI adoption journey. Different frameworks use different names, but I'll borrow the categories from Boston Consulting Group, since they map the stages most cleanly:

• Information Assistance
• Task Assistance
• Delegation
• Semi-Autonomous Collaboration

This isn't a suggestion that everyone should aspire to the top level. It's more a way to see how far things have come, and what's possible if you ever want to push further.

Stage 1 — Information Assistance: the Google replacement

This is where most people start. You open one of the common AI platforms — or your company's rolled out Copilot as part of Microsoft's offering — and you ask it a question. Instead of relying on Google's search results and sifting through whichever page Google thinks answers your query best, the AI fetches and summarises the answer directly. Even Google recognises this shift now, offering AI-summarised answers for "question" style searches (how do I..., what is...).

Try this: learn to write efficient prompts, and get in the habit of asking the AI to ask you clarifying questions before it answers.

Stage 2 — Task Assistance: AI helping with tasks

Here, you're handing AI odd jobs. You copy and paste some text and ask it to "summarise this in 250 words," or "convert the data in column A." You're no longer just enriching your own knowledge — you're getting AI to actually produce something for you. What you gain here is time.

Not long ago, AI wasn't reliable enough for this. Think of early self-driving cars: capable of handling an empty road, but not trustworthy once other cars entered the picture. Now, for tasks at this scale, AI is accurate enough — often as accurate as an average person, just faster, and it never gets tired.

I'd guess this is where most people currently sit on the AI journey — and it's also where a lot of the tension lives, in both education and the workforce. Many of these tasks used to require dedicated headcount, or let students "produce" content (write, report, summarise) seemingly out of thin air. It's a genuinely useful stage, but a tricky one on several fronts.

Try this: try AI on a range of different tasks. Notice what actually saves you time and what doesn't.

Stage 3 — Delegation

This is where AI starts replacing a whole person's worth of work — or lets someone without the underlying skill still achieve the outcome they're after, because they have the idea and AI supplies the execution. I built this website with AI's help, for instance. I have some basic web development skills, but nowhere near the level or speed AI operates at.

At this stage, you start to resemble a manager rather than a doer. You tell someone, "I want you to manage this event" — you set the constraints (budget, etc.), then let them run with it. You might check in occasionally, but you're not micromanaging every step ("make sure the venue's booked"). You simply say the event happens on this date, and trust the execution to follow.

This shift became possible after two major improvements: first, AI's ability to understand instructions and reliably produce real output (data, analysis, code); and second, output quality reaching — and sometimes exceeding — human quality in certain scenarios.

Try this: start customising AI to reflect your personality — your tone, your way of thinking, your way of working. Try building something you've never built before.

Stage 4 — Semi-Autonomous Collaboration

This is the territory where people and organisations start to get nervous, and the line between "useful" and "unsettling" gets blurry. At this stage, you're handing AI real autonomy. What does that look like in practice? In the earlier example, you decided when the event would happen. In a semi-autonomous setup, you simply tell the AI to "run the event" — it decides the where and when itself, based on available data, to maximise impact.

Scary? Maybe. And I think that's part of why some companies stop short of this stage entirely, while others use it to build multi-million dollar businesses with a tiny team, or even solo.

Try this: try building an AI agent (with sub-agents) that manages a full workflow from start to finish, unaided.

Final thought

I'm not advocating that everyone should aim for Stage 4. Through my own research, and after trying what various AI thought leaders and influencers are pitching, I've concluded that for most use cases: I can't sustain the rate at which credits get burned, the setup complexity is often disproportionate to the benefit, or I simply don't need the level of efficiency they're pushing. I call this "buying a Ferrari to fetch milk."

So ultimately, I think there's something worth learning at every stage. Understanding what's possible at the higher levels matters — but if you're already getting real benefit from where you are, the specific stage you're operating at shouldn't matter.