I want to describe how we'll work ten or fifteen years from now, and above all what we'll work with. For a century, our tools have stayed the same underneath: a place where you type information in and pull it back out by hand. I think that's ending. The tool that's coming won't be something you operate; it'll be an agent that does the work with you, and that you direct. And that shift changes more than how we work. It changes what we'll ask of a worker, and what we'll stop asking.
I'll use myself as the test case, because if I go all in on this now, I get to see sooner than anyone what it'll look like once everyone does. A few months ago, while we were putting together a bid, my agent concluded that we needed a consultant. I didn't open a browser to look for one: I told it what we were trying to win, and it found the candidates who fit, wrote to them in my voice, explained the work, and set up the meetings for the days that suited me. It handed me the whole thing with the research already done, as a trusted colleague would.
What stayed with me was more than the hours saved. Who decided we should look for that consultant? The agent did, because it had studied the bid. And who found that bid in the first place, and prepared it? The agent again. Finding the consultant wasn't the start of anything; it was one more link in a chain that began when I told the agent what I wanted to achieve. That's how I picture the work ahead: you say where you want to get to, and the tool gets you there.
The chain doesn't end there. With what I've taught it over months it has become what I called, in the last piece, my higher self, and it runs a whole plan from end to end: it organises it, carries it out, and puts it into operation. Teaching it this way, I've made it far better, and I'm sure it will know more in time. And the more of the work it does on its own, the more I keep asking something with no comfortable answer: if the agent decided we should look for the consultant and handed me the whole thing done, am I directing, or have I decided to let it decide? I think I'm still directing, though not because I checked each step, which I didn't. What I did was look at what it had put together and know a good decision when I saw one, because I'd spent a long time teaching it how I think and what I'm after. That's the judgement I can't hand over: not how to do the task, but whether, seeing it done, it's been done well. I'll go on directing for as long as I can tell a bad answer from a good one the moment I see it. The day it hands me something wrong and I don't notice, I'll have stopped directing without realising it.
Twenty years of changing the format
No one will escape this change, any more than anyone escapes technology today. The baker does the books in a spreadsheet and the mechanic plugs a diagnostic tool into the engine. That's already happened, and I'm not here to tell you what everyone knows.
The real change is a different one. Over the last twenty years we traded paper and filing cabinets for applications that cost hundreds of euros and bring the same old tasks together in one place. We called it a revolution. It wasn't. We changed the format, and here and there we made things simpler, but the work itself stayed the same: opening screens, filling in fields, moving information from one box to another. AI doesn't change the format, it changes the work underneath. That's the revolution, and it's only just beginning.
The tool stops being a tool
The computer, mind you, will remain our main tool. The screen and the keyboard are here to stay. If anything goes, it'll be the mouse, and I can't even see that happening: it gives you a kind of freedom on screen you wouldn't want to lose. The change isn't in the hardware; it's in what drives the machine.
At your desk there'll be an intelligence that doesn't just fetch data and move it around your screen, but actually does the work. It'll always be within reach, like the keyboard, and it'll generate, on the spot, the interface you need to see. Why should I learn a program and remember where its makers buried each function, when I can just say what I want to see and narrow it down by talking? The interface won't come fixed in advance for everyone; it'll be built around what you need, as you go. It's the old idea of the disappearing interface, taken to its logical conclusion: not a cleaner screen, but no screen to learn at all.
That's why the part of a CV where you list the tools you know will mean nothing. I can say this from the inside: most of the software I use at my company I built myself, without knowing how to code. If you can build a program without seeing a line of code, what does it matter which language it's written in? If an intelligence runs the spreadsheet for you, what does it matter that you've mastered it? Technical skill will count for less, and what will count for more is what the machine can't do: imagine what you want to achieve, and judge whether what it hands back is any good. In 'The End of Software' I argued that code was losing its value. This is the other side of the coin. The tool adapts to the person, not the person to the tool.
The end of searching
I've sat in front of a computer for as long as I can remember, and it still amazes me that we depend on a search engine. Typing a question, getting back ten links with the same content and a different logo, and clicking where you guess the answer might be, feels archaic to me. It's like going to the library and checking several places to track down a single fact.
Why would I spend my time gathering, making sense of and organising information, when I can say what I'm after and get exactly that in a paragraph? I think of the internet as a box that holds almost everything we know, where finding what you need has always been a knack few people had. A person skims a handful of links; a machine reads dozens and finds the answer for you.
That raises a big question: what happens to the internet when people no longer want to click ten links to arrive at the same place? If, instead of handing you information, an intelligence goes and finds the company that fits you best, websites as we know them may give way to markets where an agent matches each buyer with the company that suits them, on the strength of its track record. And here the question returns: if the agent picks the company for me, how will I know it chose badly, when I never see what it ruled out? What worries me most about that world is what becomes of the new companies, which will have almost nowhere left to get noticed; a good option the agent never shows me is, to my mind, a bad choice I never notice.
The blank screen
Picture the end of this road: a screen that's almost empty. You sit down and ask a single system whatever is on your mind throughout the day: what's happened today, whether there's anything worth reading in your inbox, what to do next, who's written to you. And out of that comes an answer built from everything that agent can reach, because it has both the data and the power to act. The email threads you untangle every day to reply to a client or sign off on a quote get sorted out in conversation. And a quiet risk comes with it: when the interface is built anew each time, there's no fixed place to notice that something has gone wrong.
And you'll explain things better by speaking than by typing. I wrote about this in 'The Higher Self': for an agent to work the way you do, it has to understand the why behind your decisions, not just the instruction. When you speak you open up, you reason out loud, and it learns to think like you. Don't speak to it like a machine you give orders to. Speak to it like someone you explain your reasons to, and tell it what you tell yourself while you work.
Once you can teach it everything, you no longer have to wrestle with interfaces someone else designed. You're no longer the one who gathers, digests and passes information up the chain. Your boss hands you the whole task, the two of you think it through, and you run it, as foreman of your agents. That's what I mean by committing to this fully. It isn't obeying the machine, it's directing it.
Your agent, like your email
Each of us will log in to an AI provider to work, the way you choose an email provider today. The computer maker may even ship it with the machine. And right there, where the search bar sits now, your agent will be. You'll do everything from it.
Some people ask, half joking, whether anyone will pay to talk to a machine; others warn it'll destroy jobs. The worry about jobs is the serious objection, and I think it's short-sighted. It assumes that what a person produces today is their ceiling. But output per person has climbed decade after decade without pause. People won't be hired for the tools they can operate, just as a footballer isn't signed for how he trains but for what he wins on the pitch. They'll be hired for results, for the problems they solve.
Someone will say that the judgement to tell good work from bad only comes from doing the task by hand, so anyone who never does it will never develop it. That's half true. At the start you do some of it by hand, just enough to get things wrong and learn to tell what good looks like. What disappears is not that learning, it's the repetition of the task once you've learned it. From then on, you build it by weighing what the agent delivers, not against what you would have done, but against what actually works: the client who complains, the figure that doesn't add up, the result. It's what happened with the calculator: you no longer had to do the sums by hand a thousand times, but you still had to understand the problem to know whether the answer made sense. I see it in my own company: that is how the eye forms, from what happens next, and once it has formed I catch a bad answer before it goes out, not when the client has already complained.
To anyone starting out, I'd say: the barrier to entry is dropping, so make the most of it. Build something, invent a way of working, become the best there is at solving one company's problems. Solve everything for one company, and there are thousands like it to sell the same solution to. Go for it.
What comes with it
I think almost all office work will go remote. Once the rote screen work dwindles to nothing, the daily trip to an office starts to look like something we can let go of too. AI and remote work go hand in hand; each makes the other easier. An agent keeps working when you're not standing over it, while a task you drive button by button keeps you tied to the spot. The more of the work that runs on its own, the less anyone needs to be there in person, until turning up becomes the exception. Each of us will go from being a cog in a department to owning a system.
Our dealings with one another will pass to our agents too. Why would I explain a problem to a person, who passes it to their own people, losing half of it along the way? My agent can hand the task straight to the other side's agent with the full context, and let that agent lay out for its owner what needs doing and how. It's how we already work at Hellomatik: each person's agent talks to the others', and when it matters, mine brings me the problem and the solution. What I watch for is exactly this: when one agent hands work to another, someone still has to notice if half of it goes missing on the way. There'll still be human contact, just less of it.
Computers, meanwhile, will keep getting thinner, and they can afford to, precisely because they won't be doing the heavy lifting. That intelligence needs computing power we can't put on a desk, so it'll live on the servers of whoever provides it, and we'll pay a subscription to use it. Your machine's power will matter less and less; its lightness, and the fact that it goes everywhere with you, will matter more. I expect the digital nomad's life to become ordinary, and the many narrow professions we have today to give way to a few broad ones. A person will go from being a hammer to a Swiss Army knife. Your limit won't be what you know. It'll be what you can imagine.
What's at stake
AI doesn't just do the same old work faster. It makes us better at solving problems, and so more capable, and that, person by person, pushes society forward. The person working ten years from now won't be a faster version of today's; they'll be something else: someone measured not by the tools they've mastered but by what they imagine and solve. They'll have gone from carrying out the work to running it, and they'll keep running it for as long as they still notice when their agent gets it wrong. I don't know if this is exactly the path, but it strikes me as the likeliest one, and I'd rather follow it to the end than watch from the sidelines.
The phone and the computer will end up as a single device. But I'll save that for the next piece.
Earlier essays by the author: 'The end of software', DOI 10.5281/zenodo.23468806; 'The machine of the future', DOI 10.5281/zenodo.22671771; 'The higher self', DOI 10.5281/zenodo.23663934.
Cite as: Correa, Isaac. 2026. "The Last Tool." Essay, October 29, 2026. https://isaac-correa.github.io/the-last-tool/. https://doi.org/10.5281/zenodo.23762228