So What Can AI Actually Do?

This is not a demo list. This is what people actually hand to AI every day.

Reading long things. Contracts, annual reports, notes from a three hour meeting, a pile of survey responses.

The question is what makes the difference. Ask it to “summarise this” and you almost always get something bland. What earns its keep is a specific question: which section covers late delivery penalties, who promised what, which numbers changed since last month's version, is there a clause that hurts us if we want to walk away halfway through.

Turning a mess into something usable. Meeting notes you typed in a hurry, converted into a task list with names and dates attached. Three thousand rows of inconsistently formatted data, normalised. A recording turned into a transcript, then into themes. Two hundred customer complaints sorted into ten problems.

This is the kind of work that used to cost an entire afternoon, and nobody has ever enjoyed doing it.

First drafts. Proposals, difficult emails, replies to complaints, presentation outlines, job descriptions, training material.

The output is rarely usable as is, and in my view it should not be. But staring at a blank page is the part that eats the most time, and that part does disappear. Our brains are far quicker at correcting than at inventing.

Explaining things again. Paste the paragraph from IT or legal that you did not follow, and ask for it in plain language. Then ask one more thing: if I agree to this, what is the risk to me.

This is the most common use I see, and the one people mention least, because nobody wants to admit they only half understood a document they signed.

Language. Translation that does not read stiffly, in both directions, tone included. The same source material becomes a formal email to a client in Tokyo and a casual Instagram caption. If your work crosses borders, this alone changes the shape of your day.

Thinking out loud with something. This is the least discussed use and in my view the most valuable. You have a decision that is bothering you, so you write out the situation at length, then ask it to build the case for why your plan will fail.

The point is not to let it decide. The point is to find the hole in your plan before your boss finds it.

If you have tried it and came away unimpressed, I can almost guess how it went.

You opened ChatGPT, typed one short sentence, got back something that sounded tidy and said nothing, and closed the tab thinking “that's it?”

The problem is how we treat it. Most people use AI like a search engine: one short question, hoping for one correct answer. It behaves much more like an intern who is fast, tireless, willing, and started this morning. It knows nothing about your office.

You would not tell a new intern to “write the proposal” and then send the result straight to a client. You would give them context first, a proposal that worked last time, who the client is, what they care about, the format your company uses. Then you would correct it twice.

Treat AI exactly like that and the output changes completely. Three things matter most:

• Material. Paste the actual document, an example of output you consider good, the raw data. Do not expect it to guess what is in your head.
• Context. Who it is for, who will read it, what tone, what length, what to avoid.
• A second pass. The first answer is raw material. Say which part is weak and ask for a fix. Two or three rounds is normal.

A quick example. “Write a follow up email to the client” will almost certainly produce something generic that could have been sent by anyone. Now compare it with this: here is the client's last email, here is what I replied two weeks ago, here is what I actually want, write a follow up that stays polite but makes clear we need a decision this week.

Same tool, same minute of your time, completely different output.

It sounds trivial. That is also the entire distance between the people who find AI unremarkable and the people who will not go back.

Numbers and sources. This is the dangerous one, precisely because it is the most convincing. Ask for data or references and it will produce them complete with a plausible title, an author who really exists, and a year that fits.

One team asked thirteen AI systems for hundreds of thousands of academic references. Half of them had never existed.^[1] Not a typo, not the wrong page. The papers were simply never written.

My working rule: anything that looks like a number, a name, a date, or a quotation is wrong until you open the source yourself.

Your office context. It does not know the client already complained twice last month. It does not know the person you mentioned is on leave. It does not know why that project was paused, or that the reason is political and should not be written down anywhere. It knows what you pasted on the screen, and nothing else.

It agrees with you. This gets discussed least and causes the most trouble. Tell it your plan is good and it will tend to agree. Flip the question to “why will this plan fail” and it will argue that just as convincingly.

It is better at pleasing than at judging. So do not use it as a judge. Use it as an opponent you have instructed to attack.

Anything that carries responsibility. There are paid AI products built for specific professions, by large companies that genuinely know the field. When independently tested, even those still returned wrong answers on a share of the questions.

If a product that specialised still misses, a general chatbot is not where you put a decision you personally have to answer for.

This is the part people discuss least, and in my view the part that matters most to you.

A software company gave AI assistants to 5,172 of its customer support staff, and a full year of the resulting data was studied by economists at Stanford and MIT.^[2]

The newest people, the ones who normally took longest to resolve a complaint, got about 36 percent faster.

The most senior people, the ones who were already good, barely changed at all.

New staff with AI closed as many cases as staff who had been doing the job for months.

So if you already write a clean email, ChatGPT feels unremarkable. If you have been rewriting the same paragraph five times, it feels like magic. Both are honest reports, from two people standing in genuinely different places.

But there is something in that number that is easy to misread.

The study measured a task with a low ceiling: closing one customer complaint. However fast you get, one ticket is still one ticket. No wonder the people who were already good could not climb much higher there.

Open ended work is a different story.

AI is very good at producing options and very bad at choosing between them. It will hand you ten versions in five minutes. Seeing that version seven is the right one, and the other nine are rubbish, takes somebody who already knows what good looks like.

That is exactly what an experienced person brings. The judgment and taste built up over years now come with a tool that produces ten options every time they ask.

So if you are senior and AI feels unremarkable, the most likely explanation is that you are still using it as a faster typist. The difference shows up when you move it somewhere else. Onto the work you keep postponing because it takes too long. Onto five versions of a strategy, built at once and compared side by side. Onto the dataset nobody ever had time to open. Onto the prototype that normally waits in another team's queue.

That is where the number goes far past 36 percent.

So both are true. The floor comes up for people starting out. The ceiling comes up too, for people who already have taste and are willing to change how they work.

This is the most common reaction I hear, and it usually comes out almost word for word:

“My job isn't really affected. I'm not worried about it.”

I understand why. If your work is meeting people, negotiating, being on site, or running a team, there is no AI that replaces that. There still isn't.

But take your job apart first.

Almost no job is one thing. It is usually twenty things, and only two or three of them are what you are genuinely paid for. The rest is supporting work: reports, emails, meeting notes, summaries, documents, explaining things to other people.

The supporting work is what just got cheap. Look at any industry:

• Restaurants. Replying to reviews, building shift rosters, working out food cost, training material for new staff, menu descriptions.
• Construction. Daily site reports, vendor correspondence, tender documents, progress summaries, site meeting notes.
• Clinics. Referral letters, record summaries, patient education material, insurance claim paperwork.
• Schools. Exam questions, report card comments, lesson material, letters to parents.
• Retail. Product descriptions, replying to customer messages, weekly sales summaries.
• Factories. Standard operating procedures, production reports, incident write ups, audit paperwork.

None of those is an “AI job”. All of them contain a pile of reading, writing, tidying and checking.

And that is the bigger picture. Every industry, without exception, runs on intelligence somewhere. Something has to be read and understood. Something has to be weighed and decided. Something has to be explained again to somebody else. All of that used to be expensive, because only trained people could do it, and trained people are in limited supply.

Most of it can now be bought for the price of a monthly subscription.

Your job is not necessarily going anywhere. What is more likely is that the person beside you doing the same work now has an extra hour every week. Or your junior, who used to need two years to reach your level, now needs one.

So if you feel safe, that is reasonable. Just check the reason again.

Because the saving is small, and it disappears into the day.

Denmark measured this carefully. 25,000 workers, in the eleven occupations that touch AI most, linked to official payroll records. On average they saved 2.8 percent of their time, about an hour a week. Effect on pay and hours worked: zero.^[3]

An hour a week is real. Nobody has ever been given a raise for an hour.

My guess at the reason is simple. Most of us use it for small self contained jobs. One email, one summary, one draft. The time saved gets absorbed straight back into meetings, notifications, and whatever else was already queued.

Not a single step of the work actually gets removed. The weekly meeting still happens. The report is still produced in the same format, read by the same people, approved through the same chain. One thing changed: the report now gets finished faster.

The teams that genuinely feel a difference have usually done the more uncomfortable thing first, which is to sit down and ask which steps no longer need to exist. That is a political question rather than a technical one, and that is exactly why it gets asked so much less often.

For this question I keep going back to one story from the history of electricity.

In 1882 Edison switched on his first commercial power station in New York. Seventeen years later, electric motors still drove less than 5 percent of the machinery in American factories.^[4]

The problem was the factory. Back then a single enormous steam engine turned one long shaft running the length of the ceiling, and every machine on the floor hung off that shaft by a belt. Which meant every machine had to stand near the shaft, and the machines were ordered by where the shaft ran instead of by the order of the work.

When electricity arrived, factories pulled out the steam engine and bolted a large electric motor into exactly the same spot. The shaft and the belts stayed hanging where they were. New power, same old factory, and almost no change in output.

It took about another twenty years for someone to notice that if every machine had its own motor, machines could go anywhere. The floor could be laid out around the order of the work. Walls could move. Roofs could open up, because there was no longer a shaft that had to hang from them.

Once factories were rearranged, the rate of productivity growth in American manufacturing more than doubled.

Forty years between that first power station and the numbers actually moving. Most of that time went on one thing: people fitting new machines into an old frame.

My view is that we are still at the pulling out the steam engine stage. AI gets used to speed up the steps that already exist, which is reasonable, because that is the easiest and least painful thing to do with it.

So my answer to “does it actually matter” is yes, slowly, and without a single big announcement. What changes first is usually small, until one day somebody rearranges their factory floor and the gap becomes visible.

ChatGPT opened to the public in November 2022. We are in year four.

The fastest way to stop guessing, and the thing I suggest to anyone who asks:

Pick one job you hate. Something repetitive, time consuming, and not a big decision. Cleaning up data, summarising a long document, turning meeting notes into a task list, writing the first draft you keep putting off.

Give it material, not just an instruction. Paste the actual document. Paste an example of output you think is good. Say who it is for and who will read it.

Do not stop at the first answer. Say which part is weak and ask for a fix. Two or three rounds is normal, and the second round is usually where it starts looking good.

Check the output line by line, once, properly. Numbers, names and dates especially. If something is off, you now know exactly where the limit is. If nothing is off, you know that too.

Do that once, seriously, and you will have your own answer to the question in the title. It will be far more useful than anything crossing your feed this week, this article included.