2025 research from DORA found that AI has made 80 percent of tech professionals feel more productive, while improving code quality for over 59 percent of respondents.
But like any advanced tool, AI’s impact comes down to the experts wielding it.
Such is the case at healthtech company Grow Therapy, where engineers are using agentic AI to connect widespread data. The result is better experimentation and faster launches.
“Our team can ship a change, read what the data says it did and use that to refine the product,” Nick Zelei, staff software engineer, said.
And the best part? It’s leading to better products for customers.
“Engineers spend less time on setup and boilerplate and more time building products that connect patients with therapists,” Zelei said.
On enterprise-size teams, like at banking and fintech company Capital One, AI has improved code generation and requirement alignment, therefore changing the pace of work.
“Instead of conducting large-scale planning exercises to fit fixed feature blocks into rigid timeframes, we transitioned to an ‘always-on’ continuous planning model,” Michael Lenner, vice president of engineering of risk tech, said.
Under this new framework, ideas can originate from anywhere, and undergo rapid, low-cost prototyping and documentation, flowing directly into a continuously refined backlog ready for delivery.
We spoke with Lenner and Zelei about how the two teams have used AI to re-shape their work and better serve customers.
Capital One, is a bank that thinks and works like a tech company, using digital fluency to improve the customer experience.
How are you using AI in your stack — and what part of that work most reflects your company’s innovation mindset?
Capital One operates with a core platform mindset. Rather than building technology for isolated use cases, we create paved paths that engineering teams across various business processes can leverage. This centralized approach drives velocity, delivers economies of scale and ensures robust risk management.
Through our AI platform approach, engineers gain access to top-tier open-weight models embedded with proper governance and controls. By expanding our engineers’ technical toolkit, we unlock innovative solutions and generate a surge of fresh ideas for complex business problems.
Specifically, my team focuses on combating financial crime. We ingest and analyze vast volumes of data to surface critical insights that enable Capital One to prevent, detect and thwart bad actors. Integrating LLMs and agentic workflows into investigation detection, research and disposition represents a true shift in how effectively we can disrupt and prevent fraudulent behavior.
What has changed in your workflow? Describe AI’s impact on your team’s speed, review and experimentation.
The most significant shift in our workflow has been the introduction of “cobuild” sessions. Throughout these daylong sessions, our tech, product and design teams collaborate directly with business partners to conduct rapid prototyping. We arrive with multiple options tailored to our understanding of the business challenge. Presenting interactive prototypes directly to stakeholders bypasses the usual distortion layer of requirements and user stories, immediately aligning everyone on the same page. Furthermore, gathering live feedback in the room and building real-time iterations allows us to arrive rapidly at a successful solution.
What impact has AI made on team delivery or customer experience?
We have addressed two primary bottlenecks that historically slowed down our delivery process: code generation and requirement alignment. For coding, we combine AI tools with our internal context and guardrails to write quality code faster. For requirements, we pair rapid stakeholder prototyping with AI-powered spec authoring to shorten the cycle for establishing high-fidelity specifications.
By eliminating these traditional constraints, we fundamentally transformed our approach to delivery and planning. Instead of conducting large-scale planning exercises to fit fixed feature blocks into rigid timeframes, we transitioned to an "always-on" continuous planning model. Under this new framework, ideas can originate from anywhere, undergo rapid, low-cost prototyping and documentation, and flow into a continuously refined backlog ready for delivery.
Grow Therapy is a healthtech company that helps independent therapists launch and grow in-network private practices through software that assists with credentialing, referrals, billing, insurance claims and more.
How are you using AI in your stack — and what part of that work most reflects your company’s innovation mindset?
I’m on the developer experience team, which exists to help engineers move faster with AI. I use it at every level of the software development lifecycle. For feature work, I reach for Cursor cloud agents, which run in a per-session sandbox behind a firewall. For other work, I run Claude Code in our own remote workspaces, also firewalled. That isolation is the point: Because we control egress and have governance in place, we can let agents run in more permissive modes without losing control.
The part that best reflects our mindset is what we’re building on top of that. Grow is at the start of defining its software factory. We adopted a durable workflow system (Temporal) as the foundation for an automations platform, so teams can define repeatable workflows, like fixing bugs and stay focused on bigger-ticket work instead of letting bug fixes delay the roadmap. The DevEx team championed this and rolled it out org-wide, and it keeps pushing us to find the limits of what cloud agents can do. That reflects how Grow approaches innovation: Engineers get room to experiment and when something proves out, we invest in the infrastructure to bring it to every team.
What has changed in your workflow? Describe AI’s impact on your team’s speed, review and experimentation.
Speed: The backlog has shrunk and grown. Items spend less time in it, but we take on work that used to sit at the bottom of the priority list. If I have an idea before a meeting, I can get an agent started before I join.
Review: Cloud agents removed the bottleneck of writing code and moved it to review. Next, we’re building review agents that give changes a first pass, so engineers can iterate before the final human review. Our sandboxing and egress controls make that possible safely. We’ll track review rounds per pull request and how often engineers act on agent feedback and refine them.
Experimentation: I use AI daily for docs, design reviews and prototyping. Two habits have changed the most for me. I point agents at documentation and ask them bespoke questions instead of browsing pages, and I dictate to agents, speaking in plain English instead of typing.
The biggest shift is owning skills for the product areas in the repos I maintain. Others aren’t experts in my area, but if I craft the right skills, their agents are more effective and need less iteration at review time. We run evals on our skills, so as models evolve and skills get refined, we can measure that they still work as expected.
What impact has AI made on team delivery or customer experience?
Our platform work reduces friction across engineering: faster setup, less toil and safe AI tooling adoption. This summer, we rolled out cloud agents to a set of beta teams, and adoption expanded. Teams overwhelmingly preferred cloud agents over local ones, and weekly merged agent pull requests increased.
AI has also tightened our quality loop. Bug fixing was a strong use case: Engineers and product managers can trigger background agents from Slack to fix bugs or make small improvements, with reviewable pull requests ready in minutes. Small issues get resolved instead of piling up in a backlog. The second shift is in how we understand what we ship. Agents can connect our data warehouse, observability and experimentation systems to investigate launches or experiments, surfacing findings that once took days to assemble. Our team can ship a change, read what the data says it did and use that to refine the product.
For our mental health care platform, the customer-facing payoff is faster shipping. Engineers spend less time on setup and boilerplate and more time building products that connect patients with therapists. We work closely with security so developers can do that without compromising privacy.



