Your product didn’t get worse. It still does the job, works the way your customers expect and gives them no compelling reason to switch. Then those customers start using agents elsewhere, and the standard around your product changes. They ask an agent to research a trip or organize some files. They use it to make a change in another product. Jobs that used to require opening an app and completing a workflow across several screens turn into a sentence. Eventually, they return to your product. They open it, find the right screen and fill in the fields, just as they have done for years. For the first time, it feels tedious because their expectations got better.
Problems customers didn’t know they had
We tend to think of customer problems as things waiting to be discovered. Talk to customers, observe their work, find the friction and solve it. But customers can be perfectly content with something until they experience an alternative. You might submit expenses manually for years without considering it a problem. You open the expense app, create a report and enter the details. That’s simply how expenses work until you start using agents elsewhere and here are my receipts, submit my expenses feels entirely reasonable. The ten minutes you used to spend haven’t become any longer. The expense product hasn’t removed a feature or made its interface worse. And yet, something you accepted as part of the job now looks like unnecessary work.
It turns out, the problem existed all along, but you had no reason to perceive it as one. Agents can expose that kind of problem by changing what customers experience as possible. Satisfaction can remain high while the standard by which customers judge the work quietly moves.
Don’t start with an MCP server
It’s tempting to jump from this observation to a familiar conclusion: agents are coming, so make everything agentic. Ship an MCP server or publish a skill. But that’s just chasing technology rather than managing a product. The broader signal matters more: are agents becoming part of how your customers get work done? If they are, their expectations may be changing before they ask you for anything. You need to understand the shift before choosing what to build.
If your target audience doesn’t use agents, there is little reason to assume they suddenly need your product to work with one. But don’t confuse that with customers not using agents with your product. They might already use agents for research or coding while they continue to operate your product manually. Maybe they prefer it that way. Or maybe you’re blocking them. Looking only for agent use around your own product creates a convenient circular argument: customers don’t use agents with us, so there is no demand for agent support; we don’t enable agents, so customers cannot use them with us.
Ask customers about the work
There is no need to invent the future. If you have a product, you should already be talking to customers. Keep asking the questions you would have asked before instead of replacing good customer research with Would you like to use AI to automate this? Ask them to take you back to the last time they did the job. What did they do, what worked and where did they get stuck? Questions about real work reveal behavior without asking customers to design the solution for you.
If agents matter, they’ll start appearing in those stories. Someone might have given a job to an agent and discovered that your product couldn’t participate. They might have taken over halfway through or found a workaround. Perhaps they decided not to try again. All this counts as evidence, and you don’t need to wait until someone cancels their subscription to take it seriously. In fact, you probably shouldn’t. Churn tells you that the change has already happened.
A happy customer can still be a warning
Agents don’t necessarily make customers less satisfied. Imagine the expense system everyone in your company is required to use. You don’t particularly like it, but now your agent can operate it for you: you hand over the receipts, the expenses get submitted and you barely think about the product. Your satisfaction with the job has increased while your awareness of the product has decreased.
For the vendor, that can look pretty good. The customer is happy while product usage and renewal remain healthy. When the renewal invoice arrives, you might simply pay it because the setup works. Unless the price is excessive, why disturb it? And even if it is expensive, switching still has a cost, and agents don’t make inertia disappear.
The warning is that the customer is becoming attached to the working setup rather than the product. They don’t care how pleasant the navigation is because they no longer navigate it. The form you redesigned and other things that once differentiated the product may disappear from their experience. So your position can become easier to attack long before you lose the customer. Satisfaction and renewal alone won’t tell you how much of the relationship still belongs to your product. They show that the current setup still works, not why the customer continues to choose it.
Make the product disappear and keep the value
Products shouldn’t fight to keep customers inside their interfaces. If customers increasingly want agents to perform a job, forcing that job through a human interface doesn’t create differentiation. Products should be automatable when customer behavior shows that the audience wants to work that way. An API doesn’t settle the question. Can an agent discover that your product is relevant and figure out how to use it? Can it complete the job reliably, or does the user need to explain the workflow in detail every time? Those are Agent Experience questions. Technical compatibility doesn’t make a product good at the job. An agent needs to understand the product’s capabilities and use them efficiently when conditions change.
Excellent Agent Experience still doesn’t guarantee that customers will feel closer to your product. The better an agent becomes at getting the job done, the less the customer may need to know about what sits underneath. And that’s okay. Your goal is to make sure that even when the product disappears from view, its value doesn’t. Ask a simple question: what would become worse if customers stopped using your product? Perhaps your product enforces expense policy or makes analytics trustworthy. Those are reasons to remain part of the solution when an agent sits between you and the customer. If the answer is mostly that switching would be annoying, pay attention. Agents are getting increasingly good at annoying things.
Don’t wait for churn
It’s easy to demand stronger evidence before doing anything. Are customers actually leaving because competitors work better with agents? Maybe not, but churn is a terrible way to discover that their expectations have changed. It confirms the risk only after a customer has found a credible alternative and acted on it.
By the time the first customer leaves, someone else has built an alternative, shown that it works and completed the migration. They have also created evidence for everyone else. There is now a case study and practical migration experience. Your competitor has a reference customer, so the second migration may be easier than the first. You don’t need to wait for that evidence when you can already see the environment around your product changing.
Talk to customers and watch how they do the jobs your product exists to support. Notice what they now find tedious and where agents enter their workflows. Then look for the places where your product gets in the way. Don’t become agentic because AI is fashionable, and don’t manufacture lock-in because agents may make switching easier. Keep giving customers reasons for your product to remain part of their solution, because your product doesn’t have to get worse for customers to outgrow it. Sometimes their world just gets better.