By Jiri Kobelka, Co-founder and CEO, ClosedLoop AI
I think the most expensive thing AI will help us build in 2027 is software nobody needs.
A coding agent can make an unwanted feature cheaper to develop. It can also make the decision to go ahead feel almost free. The launch still needs marketing, sales, support and customers willing to change how they work.
It takes an average of 11.5 months for what customers want to reach a product roadmap. That is the delay I want AI to shorten. Generating the code in an afternoon helps only after someone recognizes the need and decides to act.
My prediction for 2027 is that the advantage will move to the decision made before any of that work starts: knowing what customers actually want.
The company already has the raw material
Customer conversations are the most valuable asset a company has. People explain what frustrates them on sales calls. They describe workarounds in support tickets. They tell account managers why they are considering leaving. The company pays to have all these conversations, then uses a small part of what it learns.
Less than 2% of that knowledge makes it into new products, and only 10-15% of what gets built ends up making money. I want AI to change that relationship. Getting more work through engineering is useful when the work solves a problem someone cares enough about to use the result or pay for it.
The opportunity is to bring together what customers say, what they do and what stops them from buying. A small complaint repeated across many conversations can be more useful than a polished feature request from one persuasive person.
Imagine being able to use 100% of what the company knows. I want the 2027 conversation to be about putting that knowledge to work, with its source and context attached to the decision.
A very ordinary product decision
Consider a hypothetical company selling scheduling software to small plumbing businesses. A prospect asks whether customers can move their own appointments. An existing customer tells support that someone spends the evening calling people to rearrange tomorrow’s jobs. Usage data shows staff repeatedly moving appointments by hand.
Meanwhile, the product team has another reporting dashboard on its list.
An agent that connects those conversations can put a specific problem in front of a builder: coordinating changes is creating work for the office and frustration for the customer. The original conversations show who experiences it and how they deal with it today.
That gives the builder something useful to test. Show those people a simple way for a customer to request another appointment time. Find out whether it removes the phone calls, and what approval the office still needs. A working prototype can make that conversation much more useful than another meeting about priorities.
AI can help build the prototype. Its earlier contribution, bringing the problem into view, is what gives the prototype a reason to exist.
The agent needs to show its work
The agent should bring the original evidence into the work. A builder needs to be able to open the conversation behind a recommendation and check what the customer meant.
I want to see what customers said, when they said it and how the recommendation changes if the situation has changed. A request from last year may already be solved. Something described on a call may belong to a competitor’s product. A customer asking for a feature that already exists may need a clearer screen or a better explanation.
The person making the decision also needs to know which customers are represented. Ten requests from people the company cannot serve can have a different meaning from one repeated problem in the customers it is trying to win.
Permissions belong in that workflow too. An agent should respect the same customer and account boundaries as the person using it, including when it searches older notes or recalls something from memory. An answer is more useful when the reader can inspect the evidence they are allowed to see.
The standard I want is simple: the agent should make it easier to question a recommendation. The builder can then choose which assumption to test before committing more time.
The launch should start another conversation
The scheduling example continues after the feature is released. The company now knows which prospects asked about changing appointments, which customers described the manual work and which people agreed to test the alternative.
An agent can help find those people and prepare a relevant follow-up. The sales team has a reason to restart a conversation. Customer success has a reason to ask whether the improvement actually helped.
Someone still needs to try the new feature. The team needs to see whether the evening phone calls decrease, whether customers can complete the change and whether the plumbing business wants to keep using it. Publishing the release note answers none of those questions.
If a customer gets a useful result, ask permission before turning their experience into a testimonial. A clear account of a problem solved can help the next buyer understand why the product matters to them.
The same conversations can guide what gets built, identify who will care when it ships and help establish whether it worked. That is a much better use of an expensive conversation than leaving it in a recording.
What I expect to change in 2027
I expect more companies to organize product work around a person who owns a customer problem from investigation through to the result. Call that person a product builder. Their job will include understanding the problem, using AI to explore and build possible answers, and finding out whether customers benefited.
On November 4, 2026, we’re hosting a webinar with Mews, a fast-growing B2B SaaS company valued at $2.5 billion, about how it turned its entire R&D organization into product builders. Everyone who registers will receive the recording by email.
By the end of 2027, I expect an agent that keeps track of customer needs to become a regular companion to coding agents in these teams. It will bring relevant conversations into a builder’s work while they choose a problem, explore a solution and check the result.
The practical change I expect is a different product review. Alongside a working demo, the builder can show the conversations that led to it, the assumption being tested, the people who should hear about it and the result the team intends to check after release.
Companies should be able to measure that change. Look at how often a released improvement is used by the people who needed it, whether it removes the work they described and whether a buying objection actually disappears. Those results tell a more useful story than the number of things shipped.
That gives the company a line it can follow from a customer conversation to a product change to a customer choosing to use it, stay or buy. The agent will help keep that line visible as the work moves between product, engineering, sales and customer success.
Building will keep getting easier. Knowing what deserves to be built will be worth more.
##
ABOUT THE AUTHOR
Jiri Kobelka is co-founder and CEO of ClosedLoop AI, which helps software companies use customer conversations to make product decisions. He previously co-founded Tatum, which raised $50M and reached 170,000+ developers. ClosedLoop AI: https://closedloop.sh/
