Why I believe the most valuable artificial intelligence won't demand more of our attention, but quietly give it back.

For more than ten years, I stayed with the same broadband provider. Every year, the price crept up. Some months, the service wasn't particularly good either. I knew there were probably better deals available, perhaps even a better service for less money. Yet I never did anything about it.

It wasn't because I didn't care about the money or didn't know what to do. I knew exactly what needed doing. I had to find my contract, compare providers, understand the different packages, read the small print, and probably spend an hour on the phone negotiating or arranging a switch.

The problem was finding the time to do all of that.

Between work, family and everything else that competes for my attention, there was always something more important or more interesting to do. So I kept paying the bill, telling myself I'd get around to it eventually.

I've come to realise that this is a problem I have with much of my personal finances. I know I should regularly review my insurance, check whether my savings are earning a competitive rate, look into changes to government policies that might affect me, and generally make sure I'm making sensible financial decisions.

I'm interested in finance. I understand the importance of managing money well. But understanding something and consistently finding the time to act on it are two very different things.

For years, I've wished I could have someone take care of these things for me. Someone who understood my circumstances, stayed informed, read the fine print, and took care of the tedious administrative work. Someone who could spot a problem or an opportunity before I had to go looking for it.

Not a financial dashboard. Not another budgeting app. An actual service that looks after things.

That frustration has increasingly shaped how I think about artificial intelligence and what I believe its future should look like.

I think the greatest contribution modern AI can make to our lives is to become something we barely notice.

Not another application we have to open. Not another chatbot we have to prompt. Not another notification competing for our attention.

Instead, technology that works quietly in the background, understands what matters to us, and takes care of things without requiring us to constantly supervise it.

I think of this as ambient AI, and I believe it represents one of the most important opportunities for the next generation of technology companies.

We've been thinking about AI as a destination

Much of today's conversation about artificial intelligence revolves around what we can do with AI.

We open ChatGPT or Claude, explain a problem, provide some context, and wait for an answer. We ask AI to research something, write something, analyse a document or help us make a decision.

These capabilities are extraordinary. As someone who has spent years building software and working with AI, I'm continually impressed by what these systems can do.

But there's something about the way we're using them that I think deserves more attention.

We are still doing much of the work of managing the technology that's supposed to be helping us.

We have to recognise that something needs doing, decide to do it, open the appropriate application, explain what we want, evaluate the result, and often take the necessary action ourselves.

AI might make each step faster, but the responsibility for remembering, initiating and completing the task still sits with us.

Consider a simple example.

Imagine asking an AI assistant to compare your broadband deal against what's available in your area. It might produce an excellent comparison in seconds, identify three cheaper alternatives and recommend the best option.

That's useful. But you still had to remember to ask. You still have to review the recommendation, contact the provider, navigate the switching process and make sure everything goes through correctly.

Now imagine a different experience.

Your broadband contract is approaching renewal. A service you authorised months earlier already knows this. It checks what you're paying, researches suitable alternatives, evaluates whether switching would actually benefit you, and handles the routine work within the boundaries you've set.

If your existing deal remains competitive, nothing needs to happen. If there's a worthwhile opportunity that requires your approval, it asks.

You didn't have to remember anything. You didn't have to initiate a conversation. You didn't have to spend an evening comparing packages.

The outcome is the same, or perhaps better, but the experience is fundamentally different.

The first approach helps you perform a task. The second takes responsibility for getting it done.

That distinction is where I think much of AI's future lies.

The best technology is often the technology we stop noticing

What's interesting is that the idea of invisible technology isn't particularly new.

Some of the products I value most are the ones I rarely think about.

Take cloud backups. Once configured, services such as iCloud Backup or Backblaze quietly protect our data. We don't wake up every morning thinking about which files need backing up. We expect the service to take care of that, and only really pay attention when something goes wrong or we need to recover something.

Password managers work in a similar way. Tools like 1Password and Bitwarden sit in the background, helping us generate and retrieve secure credentials when we need them. They remove a small but repetitive burden from everyday life.

The same principle is increasingly visible in our homes.

Smart thermostats such as Google Nest can learn household routines and adjust temperatures automatically. Motorised blinds can respond to daylight and temperature. Irrigation systems can use weather information to avoid watering a garden when rain is expected.

Robot vacuums can clean while we're out. Air purifiers can respond to changes in air quality. Household essentials can arrive through replenishment subscriptions without requiring another shopping trip.

None of these products is a perfect example of artificial intelligence, and many rely on fairly conventional automation. But that's precisely why I find them interesting.

They demonstrate something important about product design that predates today's excitement about large language models.

A product doesn't have to demand our attention to deliver enormous value.

In fact, sometimes the less attention it requires, the better it is.

I don't measure the value of a backup service by how often I open its application. I measure it by whether my data is safe when something goes wrong.

I don't want to check my thermostat ten times a day. I want my home to be comfortable without having to think about it.

And I don't want a clever subscription management dashboard that constantly reminds me how much I'm spending. I'd rather have a service that notices when a subscription has become unnecessary and helps me deal with it.

There's a common pattern here: the best experience is often one where the product quietly does its job and only interrupts us when our involvement genuinely matters.

What modern AI adds to this model is the potential to move beyond predefined rules.

A thermostat can follow a schedule. A Direct Debit can pay a recurring bill. A savings app can round up a purchase.

But a genuinely intelligent system could understand changing circumstances, weigh competing priorities, deal with ambiguity and decide when an established routine no longer makes sense.

That is a much more interesting proposition than simply adding a chatbot to an existing product.

From copilots to quiet stewardship

A lot of AI companies today are building copilots.

The idea is straightforward: put an intelligent assistant alongside a person and help them perform their work more efficiently.

I think copilots are useful, particularly for creative work, software development and activities where the person genuinely wants to remain actively involved.

But I suspect we're overapplying that model.

Not every activity deserves a copilot. Many activities shouldn't require a pilot in the first place.

I don't want to become exceptionally efficient at comparing household insurance policies. I don't particularly want to develop a sophisticated workflow for reviewing my energy bills. And while I enjoy learning about investing and economics, I don't want every routine financial decision to become another research project.

I want those things handled competently, with my interests and preferences taken into account.

That means moving from AI as an assistant we occasionally consult towards AI as a form of ongoing stewardship.

A service that monitors a clearly defined part of our lives, understands what a good outcome looks like, and works towards maintaining that outcome over time.

The distinction is subtle but important.

A copilot waits for instructions. A background service knows what it has been authorised to look after.

A copilot helps us make a decision. A background service can identify when a decision is needed, prepare the options and, where appropriate, carry out the agreed action.

A copilot is useful when we're paying attention. An ambient service should continue delivering value when we're not.

Of course, this doesn't mean giving an AI system unlimited authority over our lives. I'll come back to that because I think it's one of the most important design challenges.

But the broader shift is from software that helps us manage things to software that helps ensure those things are managed.

And I think that creates an enormous opportunity.

A different way to measure product value

For decades, technology companies have placed enormous importance on engagement.

Daily active users, time spent in an application, session frequency and notification interactions have all become familiar measures of product success.

For social networks and certain types of entertainment, those metrics make sense.

But I think they can be almost backwards for ambient services.

If I'm opening my personal finance app every morning to check whether my bills are being managed properly, something might be wrong with the service.

If I'm constantly receiving notifications about routine decisions that could have been handled automatically, the product may be creating work rather than removing it.

The more interesting measures would be how much time the service gives back, how many problems it prevents, how often it achieves a good outcome, and how rarely it needs to interrupt the customer.

I think of this as quiet retention: a service becoming valuable enough to remain part of someone's life without needing to compete for their attention every day.

That doesn't mean low engagement automatically indicates success. A product nobody opens because it does nothing useful is hardly a good business.

The important combination is low effort from the customer and consistently demonstrable value.

This is also where I think a significant business opportunity exists.

Much of the technology industry competes for a limited resource: our attention.

Ambient AI offers the possibility of building businesses around giving some of that attention back.

Personal finance is where I want this future most

Of all the areas where ambient AI could make a difference, personal finance is the one I keep returning to.

Partly because of my own experience, but also because I think the problem is much larger than most financial technology products acknowledge.

We've become very good at giving people access to financial information.

Modern banking applications allow us to see our balances instantly, track spending, set budgets, move money and manage investments from our phones.

These are meaningful improvements over traditional banking.

But they haven't removed the underlying responsibility of managing our finances.

We've largely made it easier to do the work ourselves.

And that raises a question I've been thinking about for some time:

What if the next generation of personal finance products wasn't designed to help us manage our money, but to help ensure our money was being managed well?

To understand why I find this compelling, think about what it takes to stay on top of an ordinary household's finances.

You have income coming in, bills going out, subscriptions renewing, insurance policies expiring, savings that could be earning more interest, debts with different interest rates, and providers regularly changing their prices and terms.

Then there are changes outside your control: government policies, tax allowances, interest rates, new financial products and changes to the services you already use.

Even when you understand all of these things individually, keeping track of them collectively takes time and attention.

And the work is never really finished.

You might find a competitive broadband deal today, but the market changes. Your insurance might be suitable this year but not next year. An attractive savings rate can become uncompetitive. A subscription you once used regularly might quietly become something you're paying for out of habit.

This is why I've always been attracted to the idea of having someone look after my finances on my behalf.

Wealthier households have long been able to employ financial advisers, accountants, assistants and family offices to manage different aspects of their financial lives.

I'm not suggesting that every household needs a sophisticated wealth management operation. But I do think there's a universal need behind these services: having someone competent and trustworthy look after the details so you don't have to manage everything yourself.

Historically, providing that level of personalised attention has been expensive because it requires human time.

AI could fundamentally change the economics.

From financial automation to financial understanding

We already have early examples of financial services that operate in the background.

Direct Debits handle regular payments. Monzo's round-up features can automatically put spare change into savings. Robo-advisors such as Wealthfront can manage portfolio rebalancing and other routine investment activities. Services such as Rocket Money have brought subscription monitoring and bill negotiation into consumer finance.

These are useful products, but much of today's financial automation operates within relatively narrow instructions.

A rule might say to transfer a fixed amount into savings every payday, or to round up every card transaction to the nearest pound.

The trouble is that financial circumstances aren't static.

A fixed savings transfer might make sense in a normal month but become inconvenient when an unusually large expense is approaching.

An account offering the highest advertised interest rate might not be the most suitable place for someone's money if access restrictions, tax implications or deposit protection matter more.

And the cheapest broadband package might be a poor choice for someone who works from home and depends on a reliable connection.

This is where contextual intelligence becomes valuable.

Imagine a financial service that understands your income patterns, upcoming commitments, savings goals, financial obligations and personal preferences.

It knows you've authorised it to maintain an emergency buffer. When there is genuinely surplus cash, it can recommend or arrange an appropriate transfer into savings. When an expensive month is approaching, it holds back.

It notices that you're paying a high interest rate on a debt while keeping money in an account earning very little. Rather than simply displaying those two numbers on a dashboard, it helps you evaluate whether changing the allocation would leave you better off.

It sees that an insurance renewal is approaching, compares the available options against the protection you actually need, and brings you a recommendation before the renewal date.

It monitors household bills and identifies when you're paying more than you need to, while accounting for service quality, contract terms and your willingness to switch.

And it pays attention to relevant policy or regulatory changes that might affect your circumstances, rather than expecting you to discover them yourself.

None of this requires predicting financial markets or making speculative investments on someone's behalf.

Much of the opportunity is in the ordinary, unglamorous work that surrounds managing money.

Reading documents. Checking eligibility. Comparing terms. Tracking deadlines. Following up with providers. Making sure something that was agreed actually happened.

These are precisely the activities that are important enough to matter, but tedious enough to keep being postponed.

The value is in taking action

There's an important distinction between knowing what someone should do and actually getting it done.

This is something my broadband experience taught me.

I didn't need another service to tell me that comparing broadband deals was sensible. I already knew that.

I needed someone to do the comparison, explain the meaningful trade-offs, and then take care of the work involved in acting on the decision.

The same applies across personal finance.

An application that tells me I could save £150 by switching providers has given me information.

A service that investigates the options, checks whether switching makes sense for my household, obtains my approval and manages the process through to completion has provided something much more valuable.

It has taken responsibility for an outcome.

That distinction is part of what led me to start building a service that handles my household bills, beginning with broadband. Not because broadband is an especially exciting area of technology, but because it's such an ordinary example of a problem people already know they ought to address.

The bigger idea is much more interesting to me than any individual bill.

I can imagine a future where each household has a trusted financial service continuously looking after the administrative and operational side of its money.

Not necessarily making every important decision independently, but making sure opportunities aren't overlooked, problems don't go unnoticed, and routine actions don't depend on someone finding a spare evening.

I think this could become one of the most meaningful applications of modern AI.

But there's a fairly significant obstacle.

The more invisible AI becomes, the more trust matters

It's one thing to let a thermostat adjust your heating. It's another to let software make decisions involving your money.

The consequences are different, and so should be the level of responsibility.

If we're going to build systems that operate continuously in the background, we need to be exceptionally careful about what they're allowed to do.

An AI that occasionally provides a poor recommendation is inconvenient. An AI that autonomously moves money, cancels important insurance coverage or signs someone into an unsuitable contract could cause real harm.

That's why I don't believe the future of ambient AI should be unrestricted autonomy.

Quite the opposite.

The less frequently a person is expected to supervise a system, the stronger its underlying safeguards need to be.

I think there are a few principles that matter here.

The first is that people must remain in control of the boundaries.

You should be able to tell a financial service what matters to you, what it can do without asking, and what requires explicit approval.

For example, you might allow it to identify better savings rates and prepare transfers, but insist that no money moves without confirmation. Or permit limited transfers while requiring a minimum current-account balance of £500.

Other decisions, such as taking on debt, changing insurance coverage or entering a long-term contract, might always require approval.

These boundaries need to be enforced by reliable systems outside the AI model itself, rather than existing only as instructions the model is expected to remember.

The second principle is transparency without noise.

If a service is handling things in the background, I don't need a notification every time it checks a price or evaluates an option.

But I do need to know what it has actually done, why it did it, and what the outcome was.

A short weekly or monthly summary could be enough for routine activity. More consequential decisions should be visible before they happen.

I imagine the ideal communication being remarkably simple:

"I handled this."

"I need your approval for this."

"You don't need to do anything."

Those three messages capture much of what I think good ambient AI should feel like.

The third principle is accountability.

People need the ability to inspect decisions, correct mistakes, override preferences and stop a service from acting on their behalf.

Where an action can be reversed, that process should be straightforward. Where it cannot, the system should recognise the consequence and seek approval before proceeding.

Trust also means being careful with personal data, being clear about commercial incentives, and recognising when an issue requires a qualified professional or human judgement rather than automated execution.

There's also a more fundamental question about whose interests these systems serve.

A financial assistant that earns more money when it persuades you to switch providers may have very different incentives from one whose job is simply to look after your finances.

If the ambition is to build something that genuinely represents the customer, those incentives matter enormously.

I don't think technical capability alone will determine which ambient AI products succeed.

I suspect the winners will be the companies that combine intelligence with reliability, good judgement, appropriate regulation and trust.

A system doesn't become trustworthy simply because it requires fewer clicks.

It earns the right to operate in the background by consistently acting within the boundaries we've given it.

The future isn't about using more AI

One thing I want to be clear about is that I don't think every application of AI should become invisible.

There will always be activities where we want to be actively involved.

I enjoy using AI to explore ideas, write software, conduct research and learn about unfamiliar subjects. In those situations, the interaction itself is part of the value.

I don't necessarily want an AI to decide what I should be curious about, or to remove me from decisions I genuinely enjoy making.

But I think we should distinguish between activities we want to spend time on and responsibilities we simply need to make sure are taken care of.

The technology industry has often treated every problem as another opportunity to create an interface.

Another app. Another dashboard. Another set of notifications. Another habit we need to develop.

I think AI gives us an opportunity to challenge that assumption.

Perhaps the next great consumer technology companies won't be the ones that persuade us to spend more time using their products.

Perhaps they'll be the ones that become so dependable, useful and well integrated into our lives that we rarely need to think about them.

A service that quietly keeps our homes comfortable.

A system that protects our information without making security another daily task.

A household assistant that ensures essential supplies arrive when they're needed.

A financial service that stays on top of our bills, savings, obligations and opportunities without requiring us to become part-time financial administrators.

The common thread isn't automation for its own sake.

It's the removal of unnecessary mental work.

And I think that matters because our attention is finite.

Every hour spent comparing utility contracts, chasing providers, dealing with forgotten renewals or navigating unnecessary administration is an hour we could have spent doing something else.

That doesn't mean every minute we save needs to be redirected towards greater productivity.

Quite the opposite.

We could spend more time with our families, pursue our interests, learn something new, or simply enjoy having one less thing to worry about.

For all the excitement surrounding AI's ability to generate content, write code and answer increasingly complex questions, I wonder whether we'll eventually look back and find that its greatest contribution was something much quieter.

Not that it made us better at doing a thousand different tasks.

But that it relieved us of the responsibility of doing so many of them in the first place.

I've spent much of my career interested in what technology makes possible. Increasingly, I'm just as interested in what technology can make unnecessary.

I don't want to become better at administering my life. I want to spend less of my life on administration.

That's the future of AI I find most exciting: not artificial intelligence demanding a bigger place in our lives, but intelligence quietly doing its job so that we can get on with living them.