AI agents are generating a lot of excitement, and it’s easy to see why. The idea that AI could do everything from finding your next car, your next home, potentially even your next job, to booking a holiday or negotiating a better deal is a compelling one.
But the first wave of consumer-facing agents has also shown that the technology doesn’t always live up to the promise.
As these systems become more capable, we need to ask a more fundamental question: just because AI can make decisions for us, does that mean we should let it? Do we trust it and can it live up to our expectations?
Used well, AI could fundamentally change how people navigate some of the most complicated decisions they make – not by replacing judgement, but by helping people cut through noise, understand their options and get to what really matters much faster.
But there is a risk that we start treating more autonomy as an end in itself. If AI can take another decision off our hands, the assumption can quickly become that it should.
For me, the real test is much simpler: does AI help us make better decisions, and can we trust the process that gets us there?
Some decisions are too important to hand over
Buying a home, choosing a car, hiring someone or deciding what comes next in our career are not simple decisions.
A home is not just a list of property filters. A job is not just a set of skills. A car is rarely chosen on price alone. There are practical considerations such as asset value, but also preferences, emotions and priorities that are much harder to quantify.
People often work out what really matters to them as they go. You may start a property search thinking you know exactly what you want, only to realise that location matters more than space. Someone looking for a new job may begin with salary and title at the top of their list, then decide that flexibility, culture or the people they will work with matter more.
That is where AI can be extremely useful: helping people work through options, understand trade-offs and focus on what is actually relevant. This is particularly valuable for decisions people may only make a handful of times in their lives, when they do not necessarily know all the right questions to ask.
AI should narrow the choice, not make it for us
For years, the basic marketplace model was simple: bring as much choice as possible into one place and give people the tools to search and filter through it.
But more choice is not always more useful. Anyone who has looked for a house, a car or a job knows that hundreds of listings can quickly become more work rather than more help.
The opportunity for AI is to change that. Someone looking for a home should be able to explain what they want in their own words: somewhere suitable for a family, within a certain commute, near green space and within a particular budget.
AI can use that information to narrow the field, compare what is available and surface options that might have slipped through traditional filters alone.
The same applies elsewhere. Someone looking for a car may care about running costs, reliability, space and how they use it day to day, rather than simply make, model and price. A candidate may be open to roles they would never have searched for by title, but which fit their skills and ambitions.
That is a meaningful shift from search towards guidance. AI can help people get closer to the option that is right for them, without pretending there is always one objectively “right” answer.
The more AI shapes our choices, the more trust matters
Once AI starts narrowing the field, it is doing more than saving time. It is starting to shape the decision itself.
If a system reduces 500 job postings to five, or thousands of cars to a shortlist, we have to believe those recommendations make sense – that the right information is being used, important options have not been missed, and the system has understood what actually matters to us. For decisions this significant, people want to know they haven’t missed a better option. AI needs to build confidence in the recommendations it makes, rather than create more second guessing.
Marketplaces sit close to real transactions and bring together seekers and listers, buyers and sellers, employers and candidates, alongside current information on pricing, availability, demand and inventory. That context can make AI far more useful, because it grounds recommendations in the reality of the decision someone is actually trying to make.
Trust is not something a launch can buy. New entrants can spend heavily on marketing, but trust is built over years of being accurate, and accountable when things go wrong. That record takes time to build, and it cannot be replicated overnight.
But the closer AI gets to the transaction, the harder the questions become. Imagine someone buying a car with the help of an AI agent, while the seller has an agent of their own. One is trying to get the best deal for the buyer, while the other is trying to get the best price for the seller. Both may be doing exactly what they have been asked to do, but that does not necessarily mean the outcome is fair. And what happens if one has access to better information than the other?
There is a big difference between a system helping someone compare options and one acting on their behalf. As agents become more capable, we will need to ask who they are acting for, what they are optimising for, what information is shaping the outcome and who is responsible when something goes wrong.
Those questions become even more important when AI is acting on both sides of a transaction, and people may not fully understand how an agreement has been reached.
High stakes require high trust
There will be plenty of situations where people are happy for AI to take the lead – routine purchases, administrative tasks, or decisions where the consequences are relatively small.
But that changes when the stakes are higher. Research from Oxford Economics and Synchrony found that 46% of US consumers would not use AI for purchases over $5,000 yet, underscoring that higher-consequence decisions demand a much higher threshold of trust.
For major financial commitments or hiring decisions, getting it wrong can have serious consequences. Making those processes faster is one thing. Making sure people can trust the outcome is another.
If AI can help people navigate high-stakes decisions in a way that is clear, fair and keeps them in control, it can go a long way towards building trust. For marketplaces, the ultimate objective should be to create value for everyone involved in a transaction, not simply to optimise the outcome for one side.
The goal should be more confident decisions, not more automation
I am very optimistic about what AI can do here.
But success should not be measured by how many decisions we can hand over to a machine. It should be measured by whether people are better informed, understand their options and feel more confident in the choices they make.
The next phase of AI should be judged less by how many decisions it can take away from us, and more by whether it helps us make the important ones more confidently.
