All Insights
Why AI Agents Require a Different Approach to Product Design
There are two ways companies can stumble when bringing AI agents to market. The first is mistaking a working prototype for a product that’s ready to launch. By now, most understand the difference.
The second, more common risk is the opposite: spending so long trying to avoid mistakes that an agentic product never reaches customers.
Hesitation is natural. People tend to weigh potential losses more heavily than equivalent gains. For software companies, that instinct has been reinforced by the time and capital historically required to develop products.
But AI has changed that calculation.
Prototypes are now much faster and less expensive to build.1 This allows companies to resolve uncertainty during development rather than before it. The result is a new approach to product design: test successive prototypes with customers and let their feedback shape the enterprise-grade products you ultimately build.
Bringing Agents to Market Faster
Consider a team debating whether customers would pay for an agent that automates a complex workflow. Traditionally, it might spend months researching the opportunity and defining the product’s scope before committing to development.
Now, the team can build an initial version and test its hypotheses directly with customers. Who would use it? Which capabilities matter most? What would they be willing to pay?
Many companies have been slow to adopt this approach and realize its commercial potential. Only a quarter of organizations have moved at least 40 percent of their AI pilots into production, and just 20 percent report that AI is driving revenue growth.2
Here’s how Vista puts AI-assisted product design into practice, across three stages of development.
Use Case Definition
Teams often begin by limiting what an agent can do. They default to familiar products, such as chatbots, or avoid more ambitious designs that give agents greater autonomy or allow multiple agents to work together.
The starting point should be defining the customer’s problem, not a predetermined product format.
Once a team has validated the problem and established customer interest, it can test different solutions against that need. A simple chatbot might be sufficient. A more autonomous agent might deliver substantially more value.
The important thing is to let evidence determine the scope, rather than allowing assumptions about development costs to settle the question in advance.
Technical Design
Data readiness presents a similar trap. A company identifies a promising use case, then decides it must clean, consolidate and transform its data before development can begin. The resulting project can take months, even when an initial version of the product needs only a fraction of that information.
A more practical approach is to work backward from the use case. Identify the data required to build and safely test an initial version, prepare that data and expand the underlying infrastructure as the product develops.
Collaborative Build
Even after a working prototype exists, teams can be reluctant to put it in front of customers. They continue refining features and optimizing performance, sometimes spending tens of thousands of dollars on improvements with an unknown commercial value.
At a certain point, customer feedback becomes more valuable than further internal refinement.
Early trials reveal which capabilities customers value and whether they’re willing to pay for the product. Teams can use that evidence to guide development and pricing before investing further in optimization.
The principle across all three stages is simple: spend less time trying to predict what will work and more time finding out.
Recalibrating Risk
None of this is an argument for loosening security, governance or quality standards. An agent that handles sensitive data or acts on behalf of customers always requires appropriate safeguards. Faster development does not make those obligations any less important.
But many enterprise software companies have advantages that allow them to develop AI products faster while managing risk:
- Established customer relationships provide access to early feedback on use cases and product design.
- Proprietary data can provide the information needed for specific use cases without requiring a comprehensive data overhaul.
- Years of deployment experience mean testing infrastructure and safeguards may already be in place, allowing teams to experiment safely without building them from scratch.
The Upside
“When I think about how we would have gotten this done without the Agentic Factory, I think it really comes back to, would we have been as bold?”
Catherine Cormier
Chief Innovation Officer at Assent, a Vista portfolio company³
The instinct holding companies back from launching AI agents predates AI itself. It’s partly bias: people tend to weigh potential losses more heavily than gains. And it’s partly habit, formed when testing a new product required months or years of work.
If time to production is no longer the same constraint, why should companies limit the scope of their ambitions? Instead of asking which single agentic feature is worth the investment, a company might ask which ten ideas are worth testing. It may discover an opportunity it would otherwise have dismissed as too ambitious.
We believe companies that embrace this approach can move faster to capture the value AI is creating for software.
Read more about how Vista’s Agentic Factory helps portfolio companies aim higher and launch agentic products faster.
