The future of AI at work, and what to focus on now

1The saree and the T-shirt

Riya's board exams are coming, so she's revising a history chapter called "The Age of Industrialisation". On her chair hangs her favourite T-shirt, made in a factory. In the cupboard is her nani's wedding saree, a Banarasi, woven by hand, thread by thread.

Why are the two so different? For thousands of years, all cloth was woven by hand. Some of it was stunning, some of it was rough, and no two pieces came out quite the same. India's cloth was among the best: her textbook says that before machines, the finest cottons and silks in the world often came from India.

But then came machines. In Britain, power looms wove faster than any person could, and they never got tired. Machine cloth was cheap, even and everywhere. So it flooded the world, India included. Riya's chapter has the numbers: in 1811–12, cloth made up 33% of everything India sold abroad. By 1850–51, it was just 3%.

That change hurt real people. Many weavers lost their living, in India and in Britain. British hand weavers earned about twice a factory worker's pay in 1806, and a quarter less than one by 1820. Any honest story about new technology has to say that part out loud.

But here's the twist in Riya's chapter. The weavers didn't disappear. Between 1900 and 1940, the handloom cloth made in India almost tripled. Two kinds of weavers did well. Some picked up a new tool, the fly shuttle, which let one weaver work much faster: by 1941, over a third of India's handlooms had one. Others wove what the machines couldn't copy: Banarasi and Baluchari saris, woven borders, intricate designs.

So the machine did two things at once. It raised the floor: suddenly everyone could afford decent, even cloth. And it moved the value up, to the things only skilled hands could do. That's why a hand-woven Banarasi still costs far more than a factory T-shirt, and why India's last handloom census still counted around 35 lakh weavers and helpers.

2A new floor for digital work

Now look at Riya's laptop. The same thing is happening to everything made on a screen. Not long ago, a school project, a sales deck or a report could be anything from beautiful to a mess. Today, anyone can type one line and get a neat deck, a tidy PDF, a decent video script or a polite email. The rough, uneven end of digital work is disappearing, just like rough hand-woven cloth did.

Is that just a feeling, or has anyone measured it? Researchers have. In a 2023 study of 5,179 customer-support agents, an AI helper raised the number of problems solved per hour by 14% on average. For the newest, least experienced agents, the jump was 34%. The most experienced agents barely changed.

Another experiment, by Harvard researchers and the consulting firm Boston Consulting Group (BCG), tested GPT-4 with 758 consultants. The ones who started below average improved by 43%. The ones who started above average improved by 17%. Everyone got better, but the bottom rose the most.

That rising bottom line is what we'll call the new floor: a basic standard of quality that every deck, PDF, email and video will have, because AI makes it almost free. Like factory cloth, it won't feel special. It'll just be everywhere.

3Your edge is the non-average

But there's a catch with a floor. When everyone stands on it, everyone looks the same. The same Harvard and BCG study found that ideas from consultants using AI were better, but noticeably more alike. Another study, from 2024, gave short-story writers ideas from AI. Their stories were rated more creative, especially for writers who weren't very creative to begin with. But the stories also came out more similar to each other.

Why does that happen? Remember, AI is a giant guesser, and its best guess is the most likely answer, which is the average answer. So when everyone asks the same AI the same thing, everyone gets roughly the same average.

Let's watch it happen. It's Kabir's birthday, and Riya asks AI to write him a message.

So where does that leave us? With everything above the floor. Taste: knowing what good looks like. Nuance: the small things that matter to this person, right now. Judgement: deciding what's true, what's right and what to do next. AI gives you the average. Your edge is the non-average.

4The jagged edge

So what can't AI do? The honest answer is: it's surprising. Picture a batter who can hit the fastest bowler in the world for six, but keeps getting out to a slow, gentle ball. You'd never guess which ball gets him out until you've watched him bat many times.

AI is that batter. It can draft a whole business plan in a minute, then miss a clue sitting right in the notes you gave it. The researchers behind the Harvard and BCG study have a name for this: a jagged frontier. Some tasks are easy for AI, while others that look just as hard are beyond it.

They tested both sides of the edge. First, a task inside it: invent a new shoe for a sport or group that's poorly served, and plan its launch. Consultants with AI finished 12.2% more of the work, 25.1% faster, and their work was rated more than 40% better.

Then a task just outside it. Advise the boss of a made-up company, using a spreadsheet and some interview notes. The spreadsheet looked complete, but a careful read of the interviews changed the answer. Without AI, 84.5% of consultants got it right. With AI, only 60 to 70% did.

Here's the scary part: the advice written with AI was rated higher quality even when its answer was wrong. AI sounds just as confident and polished when it's wrong as when it's right. So the only way to learn where its edge is: use it a lot, and check whatever matters.

5The electricity bill

Now, back to Mum's puzzle. Back in 2017, the AI researcher Andrew Ng called AI "the new electricity". It's a good picture, so let's take it seriously, starting with the bill.

How does Riya's family pay for electricity? By the unit. A meter counts every unit the house uses (one unit is a 1,000-watt heater running for an hour), and the bill is the total.

AI is billed the same way. Its units are tokens, the chunks of text an AI reads and writes. Every token you send in and every token it writes back is counted. AI companies price them per million tokens, with one rate for what goes in and another for what comes out.

And the units keep getting cheaper. Stanford's AI Index found that running an AI as good as the first ChatGPT became more than 280 times cheaper between November 2022 and October 2024. Like electricity, AI is turning into something you simply plug into.

But here's the thing about any bill. It tells you how much you used, not how much good it did. A bulb left on in an empty room costs exactly the same as one lighting up homework. So Mum's company was paying for units. The real question is: what were those units connected to?

6The 30-year puzzle

Let's travel back to a factory around 1900. In the corner sits one huge steam engine. It turns a long iron shaft that runs along the ceiling, and leather belts drop from that shaft to every machine on the floor. When the engine runs, every machine runs. When it stops, the whole factory stops.

Then electric motors arrived. So factory owners did the obvious thing: they took out the steam engine and put in a big electric motor. Same shaft, same belts, same building. And the result? Not much. In 1899, electric motors supplied less than 5% of the power in American factories, and it took about twenty more years to pass half. Even then, productivity hardly moved until the 1920s, some thirty years after the first factories went electric.

Why did such a huge invention take so long to pay off? An economist called Paul David asked exactly this in 1990, in a famous paper called "The Dynamo and the Computer". His answer: the motor wasn't the problem. The factory was.

The gains came when factories were rewired. Instead of one giant motor, every machine got its own small one. With no heavy shaft to hold up, buildings could be lighter and single-storey, with skylights. Machines could stand in the order the work actually flowed. One machine could be fixed without stopping all the others. When American factory productivity finally jumped in the 1920s, David found that about half of that jump went along with factories adding those small motors.

So here's the answer to Mum's puzzle. The power was never the problem. The wiring was. Her company plugged in AI and changed nothing else: same steps, same handoffs, same copying and pasting between apps. AI sits in a chat window on the side, like one big motor turning the old shaft.

And Mum's company isn't alone. A much-discussed 2025 report from a team at MIT found that about 95% of the company AI pilots it studied showed no measurable effect on profits. Its main explanation wasn't weak AI. It was AI tools that didn't fit into, or learn from, the way the work actually gets done.

7What wiring means for AI

So what does rewiring look like? Let's follow Mum's morning. A new customer emails. She copies the email into the AI chat, types a prompt, copies the answer back, then updates the sales records by hand. The AI is on the side. Mum is the wire.

Rewired, the AI becomes an agent that works inside her tools. Five wires make that happen:

• Connections: which apps it can reach, like email, calendar, files and the sales records. This is where MCP comes in.
• Permissions: what it may do on its own, and what it must ask about first, like sending, paying or deleting.
• Context: what it knows about her work: her customers, her templates and how her team does things.
• Privacy: company data stays inside the company, and personal notes stay personal.
• Training: people who know what it's good at, and where its jagged edge is.

Now when a new email arrives, her agent has already checked the customer's history, drafted a reply in her style and updated the records. Mum reads it, changes one line and presses send. Same AI, rewired.

8Where MnemioAI fits

This is the part where we tell you what we do, because this whole series has been quietly leading here. MnemioAI does the wiring.

We give every employee their own AI agent, connected to the tools they already use: email, chat, docs, the sales system and more. Each agent is set up for the company, with the right permissions and context, and personal notes stay private. Agents find, draft and prepare. People review and press send. And we train the team, so everyone knows what their agent is good at, and where to check its work.

If your company is paying the AI bill but still waiting for the gains, see how MnemioAI works for businesses, or book a 30-minute demo and we'll show you on your own work.

9What to focus on now

So what does all this mean for you? Whether you're 15 like Riya or running a team like her mum, the plan is the same:

• Use it every day, on real work. Revision plans, emails, reports, practice questions. Not just for fun.
• Learn its jagged edge. Notice where it slips, and check anything that matters: facts, numbers, names.
• Bring what it can't. Your context, your taste, your judgement: the inside joke, the clue hidden in the interview notes.
• Start now. As the Wharton professor Ethan Mollick puts it, assume this is the worst AI you will ever use. It will only get better, and so will the people who started early.

The World Economic Forum expects about 39% of workers' core skills to change by 2030. Skills with AI and data are growing fastest, but creative thinking, curiosity and lifelong learning are rising too. That's the floor and the edge, in one list.

Remember the weavers in Riya's chapter? The ones who did well picked up the fly shuttle, and kept weaving the borders no machine could copy. That's the whole plan. Let AI raise your floor, wire it into how you really work, and spend your time on the edge only you can weave.

And underneath all of it? An AI model still does one thing: it guesses the next word, the way you do when someone says "the sun rises in the…". Everything useful AI does comes from what we build around it, and from what we bring to it. That was true of electricity, and of the power loom, and it's true now.

One last confession. We had to make a five-part series to explain what we do. Maybe this is what they call educating the customer.