All Insights
A Tour of the Agentic Factory: How Vista Helps Portfolio Companies Launch Agentic Products
Writing code used to be the hard part of building software. Today, Agentic AI has changed that: software can be prototyped in hours and developed in days. The hardest work is now ensuring that users not only have a great experience, but that what is being built delivers real value and is genuinely useful.
Vista’s Agentic Factory was created in 2025 to help software companies navigate this complexity. Our dedicated team of experts works directly with portfolio companies across the full development lifecycle to identify, design and ship agentic products. Since launching, the Factory has been the engine helping drive the portfolio’s most ambitious AI work.
A typical engagement has three stages – use case definition, technical design and collaborative build – and can take a company from an idea to an enterprise-grade agent in days.
Stage 1: Solution Design
Every Agentic Factory engagement starts from the same principle: “a problem well-stated is a problem half solved.”
Before deciding what to build, teams need to ground their work in a pain point. Who experiences it? Why does it matter? And what would a better outcome look like?
Companies come to that conversation prepared with customer feedback, gaps in their existing products and a view of what their competitors are already shipping. The Agentic Factory brings together roles from across the business to participate, including:
- Product and engineering leads, to align on the market opportunity and audience before the build.
- Domain experts, to contribute industry expertise that generic models can’t.
- Senior leadership, to help make the decisions required to move quickly.
Together, this team translates complex use cases into a multi-step workflow, accounting for key variables. The result is a blueprint for the product showing where agents, software and humans each fit in. Stakeholders weigh the benefits of automation against risk and compliance requirements to determine how much autonomy to give each agent.
This process often leads companies to solutions they didn’t expect. One portfolio company arrived looking to build a chatbot that could configure products from specification documents running thousands of pages. After mapping the use case, they instead landed on agents that could work with files asynchronously.
This stage takes time to get right, but defining the use case and solution clearly here helps everything downstream move quickly.
Stage 2: Technical Design
The technical design stage puts the systems and data in place to bring the blueprint to life.
This isn’t as simple as handing a data packet over to an agent. Teams need to map every system the agent will interact with, including where it gets information, where it takes action and how its performance is monitored.
Data readiness at this stage is about precision over volume. Companies should pinpoint and prepare the relevant datasets needed for the scoped use case, rather than running a months-long data transformation upfront.
Evaluation datasets are typically the toughest piece to solve. Agents are probabilistic: the same task can produce different outputs, and a product team needs a mechanism to distinguish an accurate result from a bad one. Most companies do not already have evaluation datasets built for this purpose.
Consider Accelya, a Vista portfolio company and a leading software provider to the airline industry. The Agentic Factory worked with Accelya to develop agents to support complex airline workflows, such as disruption management, order intelligence and settlement decisions. For an agent recommending an action during a disruption, for example, the team needs a set of representative scenarios and expected outcomes against which its recommendations can be evaluated. That allows the team to test whether the agent is consistently making appropriate decisions before deploying it into live operations.
Creating these evaluation datasets begins during technical design and continues into product iteration. This is how teams monitor the accuracy of agentic work during development as well as when the features are released at scale.
Stage 3: Collaborative Build
Finally, it’s time to build.
AI has dramatically expanded what a hackathon-style sprint can accomplish. Teams can have a working MVP (minimum viable product) by the end of day one and an enterprise-grade agent in as little as three days. The Agentic Factory did just that with supply chain management solution Assent, compressing what used to take several multi-week sprints into a few days.
Faster coding explains only part of that acceleration. Traditional software development loses enormous amounts of time between decisions. Engineering builds something. Product reviews it. Leadership weighs in. Feedback travels back to engineering. A relatively small change can spend days or weeks moving through the organization.
The Factory compresses that loop by putting decision makers in the room with engineers and domain experts. Pivots that once took days or weeks of realignment now happen in hours.
The Vista Model
Nearly a dozen portfolio companies have gone through the Agentic Factory so far, spanning industries including insurance, fintech and supply chain.1 Among them:
- Duck Creek, an insurance solutions company, designed a set of coordinated First Notice of Loss (FNOL) agents that now help process more than $60 billion in insurance premiums annually.2
- Finastra, a global fintech company, built its OperatorAssist tool to speed up payments exception handling, and it’s delivered efficiency gains upward of 20 percent since launching earlier this year.3
- Accelya, an airline software company, brought agents to flight disruption workflows across a platform that serves more than 200 airlines worldwide.4
- Assent, a supply chain compliance software company, worked with the Agentic Factory to deploy agents that support proactive risk identification and guided remediation.
Each engagement moved from concept to enterprise-grade product in days instead of the months or years once required.
That speed is proof of Vista’s Operational Intelligence℠ model at work. The Agentic Factory itself is a small team that runs on the same agentic capabilities it builds for portfolio companies: researching, coding and stress-testing solutions with agents alongside humans. The team moves beyond simply advising on agentic products and brings its own operating model directly to portfolio companies.
Learn more about how the Agentic Factory supports Vista’s portfolio companies.
The New Rules of Building Agentic Software
