Tech Corner: How Cut+Dry Got to 30% AI Code Generation In Just a Few Months

AI, AI, AI. It's everywhere. But AI for coding was a relatively new thing a year ago. It has grown exponentially since then, like everything AI usually does. At Cut+Dry, we began experimenting with AI for coding in late 2024. As with any new technology, it started with a few early adopters. The team initially started with Microsoft Co-Pilot and then began testing out Cursor, a new AI code editor tool. 

Our engineering leadership connected with a Silicon Valley CTO early in 2025 to understand how their company was using AI coding tools. Cut+Dry’s leadership was very supportive and encouraged the adoption of AI for coding after seeing multiple demos. After thoroughly evaluating the privacy policy and data usage terms to protect our proprietary data, we began a 15-engineer pilot program across multiple teams. The pilot went on for 2 weeks, and feedback was overwhelmingly positive. Engineers were excited about AI and the possible productivity gains and championed a company-wide pilot. 

AI Coding: Bottom-Up Adoption

As I said before, early adopters were key to adopting AI in coding. We identified these early adopters from each team to serve as unofficial AI leads. We deliberately avoided forcing AI adoption. It was important to protect Cut+Dry’s work culture, where we encourage engineers to follow a strategy they feel comfortable with. 

We decided to proceed with Cursor and extended the pilot program to all software engineers. We did not have any formal training; instead, we relied on self-learning, online resources, and knowledge sharing. We had weekly engineering sync-ups, including demos and discussions of successful use cases to encourage adoption among the engineering team. 

We did not use tracking tools or metrics that could create pressure. We focused on showcasing the benefits of using the tools and demonstrated successful use cases. We wanted engineers to buy into it themselves rather than being forced into it. We prioritized organic adoption over a top-down mandate. 

Key use cases for AI coding at Cut+Dry

Throughout the pilot process and eventual adoption, we identified several key use cases for AI coding at Cut+Dry. We could not use AI coding for everything on Cut+Dry’s legacy code. It's easy to use AI coding when you are starting from scratch because it’s a clean slate - but with legacy code like we had, AI coding is best used for specific cases. Some of the key use cases are listed below:

• API integrations emerged as a major use case, reducing development time from 1-2 weeks to 3 hours for new connectors. 

• UI creation accelerated by using AI with Figma mockups to generate front-end code.

• New engineer onboarding improved as AI helps explain legacy codebase without requiring a senior engineer's time.

• Product team members now use AI to write simple scripts for data cleanup tasks, getting senior engineer's review before execution.

• AI serves as a "senior engineer" for answering questions about the codebase, handling approximately 75% of queries.

This basically helped us get to 30% of all new code written by AI within a few months; the product team could complete certain tasks without waiting for engineers' availability, new engineers onboarded faster by using AI to understand the codebase. Although we had no formal measurement system, anecdotal evidence shows significant time savings for our teams. 

Challenges and pain points with AI code generation

AI code generation is not without its challenges. As a startup, we must analyze the costs and benefits of using tools like these. Although it seems like a significant expense, when we look at the productivity gains we got from using AI, the benefit far outweighs the cost. 

AI-generated code still requires thorough review as it doesn't always match company standards; it doesn't write code the same way as an experienced engineer familiar with the codebase would. Initially, we had no built-in way to track which code was AI-generated versus human-written, so this was a challenge for reviewers. We adapted to this by requiring engineers to indicate AI-generated code in pull requests so that the reviewers could pay closer attention to it. 

Another key challenge was measuring our performance in AI code generation adoption and setting milestones for the future. Since this is still a new area, we did not find existing maturity models specifically for AI code generation. So we had to create our own model. 

We created a 5-level maturity model to guide implementation. It helps track progress, set boundaries, and establish clear goals for advancement. This was important because having a framework in place will prevent scattered, unorganized adoption across different teams and provide a structure for measuring progress and planning next steps. 

Conclusion

Despite the sizable cost, the investment has been worth it for Cut+Dry. Being able to generate 30% of all new code using AI is a major game-changer for a start-up like ours. AI coding technology too is growing rapidly; it has grown so much in just the few months since we began experimenting. In my opinion, our bottom-up approach was key to the successful adoption of AI in coding at Cut+Dry. It was like a seed that we planted and nurtured - now it's growing into a mighty tree.

---

This blog was written by Krishan Senevirathne, Vice President of Engineering at Cut+Dry. Krishan has over 10 years of work experience in the software engineering industry. He started his career in 2011 at IronOne Technologies LLC. He joined Sysco Labs in 2013 where he went on to become a Senior Technical Lead. He joined Cut+Dry in 2021.