The Last Generation That Had to Learn How to Use Computers

I sometimes wonder whether we're living through the last period in human history when being good with computers means knowing how to use computers.

For most of my life, technological competence has been measured by how fluently you navigate someone else's interface. Can you find the right menu? Understand a spreadsheet? Set up a database? Remember which application contains the thing you're looking for? Know which button to press without accidentally emailing 4,000 people?

We got so used to this arrangement that we stopped questioning it. Humans have spent decades learning the language of machines. What happens when the machines finally learn ours?

The enormous effort of learning to use a computer

Consider how much of modern working life consists of interacting with software rather than accomplishing the thing the software supposedly exists to help us do.

A salesperson doesn't want to update a CRM. They want to understand a customer and win their business. An event organiser doesn't want to master seventeen configuration screens. They want to run an exceptional event. A founder doesn't want to reconcile information across applications. They want to know whether the company is healthy and what needs their attention.

Yet we've made operating the tool a substantial part of the job. Whole training industries have emerged around making people proficient in interfaces that were designed around the constraints of databases, operating systems and software engineering.

There's nothing inherently wrong with a good interface. I care enormously about design. But we should recognise the peculiar bargain we've accepted: we learn how the computer organises the world so that we can ask it to do something useful.

AI offers the possibility of reversing that bargain.

I can see the transition happening as I build Topolo

This isn't a prediction I've arrived at from watching product launches or reading breathless AI announcements. It's something I increasingly experience while building Topolo.

As the platform has grown, so has its roadmap: interconnected applications, each with its own domain, data, permissions, workflows and user experiences. On paper, that sounds like traditional enterprise software. In practice, the way I interact with it is changing.

More and more, instead of opening an application, finding a record and manually clicking through a process, I describe an outcome. Agents investigate, use the appropriate systems, attempt the work and return evidence for me to review.

The important part is that the underlying applications haven't stopped mattering. If anything, they've become more important.

The system of record remains. The obligation for a human to operate its interface starts to disappear.

A customer still needs a reliable identity. An invoice still needs an auditable financial record. A meeting still needs an accurate calendar entry. Permissions still need to be enforced, and actions still need to be reversible where possible. Telling an agent to 'sort it out' doesn't relieve us of any of those responsibilities.

What changes is the layer between the person and the underlying business systems. Instead of us learning the paths through the software, the software can increasingly work out the paths on our behalf.

Why ten disconnected apps are a bigger problem in the agent era

There is a popular argument that AI agents will make every SaaS application interchangeable. If an agent can use any API or browser, why worry about the underlying software estate?

I think that misses something fundamental.

An agent operating a business scattered across a dozen vendors inherits a dozen interpretations of the same customer, a dozen sets of permissions, a dozen integration behaviours and all the inconsistencies between them.

One system says the customer is active. Another says their subscription is overdue. A third has the latest conversation, but nothing has copied it back to the CRM. The agent may be extremely capable and still spend most of its effort trying to establish what is actually true.

A coherent platform, with interoperable systems of record, gives both humans and agents a clearer foundation. Less ambiguity. Fewer fragile connections. Better traceability from a decision to an action and its eventual outcome.

That is one of the central ideas behind Topolo: not an enormous collection of apps for the sake of having an enormous collection of apps, but an environment where the underlying capabilities belong to a consistent operational world.

Agents can work across external tools too, of course. No single platform can or should own every piece of software in existence. But integration is not free just because an AI can write the integration code. Reliability, provenance, permissions and recovery still matter.

The interface may become disposable

Here's the consequence I find even more exciting: the economics of creating interfaces have changed.

Traditionally, designing a beautiful, functional interface was a significant investment. You created a set of screens, shipped them and hoped that thousands of people would use them repeatedly enough to justify the work.

But increasingly capable models can now help generate an interface for a particular problem, person or decision. Not just the copy inside a reusable template, but potentially the whole interaction.

Imagine asking: 'Show me which of our enquiries from the last month could become valuable long-term customers, explain your reasoning, and let me choose which ones deserve a personal follow-up.'

Why should the answer necessarily be a fixed CRM screen built three years ago? It could be a purpose-built visual workspace with the right evidence, comparisons and controls for that exact question. You might use it for five minutes and never open it again.

Tomorrow you might ask for a completely different view: supplier risk, a proposed event budget, a comparison of alternative pricing models, or the dependencies stopping a product release.

Each could have its own beautifully designed interface. Some might only ever be seen once.

Permanent data. Durable business logic. Temporary interfaces.

That doesn't mean design is irrelevant. It could mean design becomes much more pervasive. We may need extraordinary consistency in typography, accessibility, brand language, interaction patterns and trust signals even when the composition itself changes every time.

A generated interface that looks wonderful but hides an irreversible action is not progress. A beautiful chart with incorrect numbers is merely a more persuasive mistake.

The future I want isn't an endless stream of improvised screens. It's a system capable of composing the right interface on demand while preserving the safeguards and visual discipline of a serious product.

The new job of the human interface

If agents increasingly do the clicking, what are interfaces for?

I suspect they'll shift from being primarily places where we perform work to places where we understand, direct and verify work.

Instead of entering every field in a workflow, the human decides whether the proposed outcome is sensible. Instead of navigating a sequence of screens to build a report, they question the assumptions behind it. Instead of doing thirty routine steps, they approve the one consequential choice.

This is why I remain so interested in interface design even as I work to reduce the amount of human attention required to operate Topolo. Those two goals aren't contradictory.

The more execution we delegate, the more important it becomes to present complex work in a form that a person can understand quickly and accurately.

My own bottleneck is increasingly not the quantity of work agents can attempt. It's the quantity of completed work I can meaningfully evaluate, approve or redirect. I've started thinking about productivity as completed, verified work per minute of human attention.

That requires interfaces which are excellent at revealing uncertainty, communicating evidence, showing meaningful differences and making the next human decision obvious. Less menu archaeology. More judgement.

Who becomes 'good with computers' next?

For decades, technical fluency has been a powerful advantage. People who could programme, master complicated spreadsheets or navigate enterprise systems could do things that others simply couldn't.

AI has the potential to make many of those capabilities much more accessible. Someone who knows exactly what outcome they want may increasingly be able to get there without learning all the machinery underneath.

But I don't think that means expertise stops mattering. It changes where expertise creates value.

Understanding a customer's real problem, recognising a bad assumption, designing a sensible business model, exercising taste, identifying a risk or deciding what should exist in the first place—none of those become trivial simply because software is easier to operate.

In fact, people who combine domain knowledge, imagination and judgement with the ability to direct intelligent systems may become extraordinarily capable.

There will also be new inequalities. Access to reliable systems, good data, trust and the ability to verify what an agent has actually done may matter as much as access to the model itself. 'Anyone can build anything' makes for a wonderful headline, but reality still has customers, regulations, cash flow and consequences.

The old technical advantage may shrink in some places and become more powerful in others. What changes is the entry point.

A generation that might never need to learn the menu

I remember the days when knowing where files lived, how to configure a machine and which commands to run felt like the difference between being capable with technology and being lost.

Imagine a generation for whom that distinction barely exists. They express a goal, receive a sensible proposal, inspect the result and decide what to do next. A document, an application, a database or a temporary interface might be produced along the way—but they don't have to know in advance which piece of software is supposed to contain the answer.

That future won't arrive everywhere simultaneously. Legacy systems are stubborn, some tasks require precise direct manipulation, and regulated environments demand clear accountability. There will always be good reasons for stable, familiar interfaces.

But I doubt the default expectation—that we must first learn the computer's language before we can make it useful—will survive unchanged.

We may even look back at the endless courses teaching us how to navigate software menus with a certain affection. Somewhere, a PowerPoint certification will presumably still be available. Old traditions die hard.

The software doesn't disappear. It finally gets out of the way.

I don't believe the future is a world without applications. It's a world in which applications become dependable capabilities rather than destinations we are constantly forced to visit.

We will still need systems that remember accurately, transact securely, respect permissions and preserve a trustworthy account of what happened. We may need them more than ever.

But the interface can become contextual, personal, temporary and, when it serves no further purpose, disposable.

A founder shouldn't have to become an expert in CRM software to understand their customers. An operator shouldn't have to learn the architecture of six tools to get one useful answer. And a brilliant idea shouldn't be held hostage by someone's familiarity with a settings menu.

For fifty years, we adapted ourselves to computers. The next chapter is about computers adapting themselves to us.

I suspect we're closer to that transition than most people realise. I'm watching it emerge through what we're building at Topolo, one workflow and one interface at a time.