In an era of rapid technological transformation, artificial intelligence is reshaping the landscape of software development. VMblog recently sat down with Dave Cole, Chief Technology Officer at Formstack, to dive deep into how AI tools are revolutionizing app development, enhancing developer efficiency, and paving the way for more intelligent, autonomous computing platforms. From automated testing to agentic workflow models, Cole offers a candid look at the promises and challenges of integrating AI into modern software engineering practices.
VMblog: What AI tools have helped us improve in-house app dev?
Dave Cole: Several AI tools have significantly boosted our in-house app development. GitHub Copilot has been useful for improving developer productivity by offering code completions, refactoring suggestions, and handling repetitive coding tasks. While it hasn’t been transformational, it’s been a reliable tool for speeding up development cycles and reducing cognitive load on developers. Another tool, AWS Bedrock, has been crucial for safely harnessing commercial, pre-trained AI models for rapid feature development. It allows us to integrate powerful capabilities, such as language models and image generation, without extensive setup. More recently, Cursor has proven to be a game-changer for increasing productivity, particularly in automated unit test generation. This tool helps write and refactor unit tests quickly, improving both our test coverage and code reliability. Overall, these tools have helped streamline various aspects of development, saving time and reducing manual effort.
VMblog: What parts of app dev have been taken over by AI?
Cole: AI has taken over some important tasks in app development, especially in test coverage. AI-powered tools can now automatically generate unit and integration tests, catching bugs earlier in the development process. This automation has allowed us to improve code quality and maintain higher test coverage with far less manual effort. On the other hand, tasks like front-end development are still mostly handled by developers. While AI can assist with some elements, such as suggesting layout structures or UI components, generating high-quality, responsive front-end code-especially from design tools like Figma to React-is still not reliable. AI can automate basic elements of UI development but is not yet capable of creating complex, interactive interfaces without human oversight. For now, the major AI impact has been on back-end development and testing, with front-end coding remaining a human-centric task that requires creativity and technical finesse.
VMblog: How did you measure developer efficiency / quality and speed after allowing teams to use AI tools?
Cole: We’ve measured the impact of AI tools on developer efficiency through key performance metrics like sprint burndown charts and test coverage statistics. Since incorporating AI into our workflows, our sprint velocities have consistently increased, with teams regularly hitting or exceeding 100% of their sprint goals. The use of tools like GitHub Copilot and Cursor has freed developers from time-consuming tasks such as boilerplate coding and manual test-writing, allowing them to focus on more complex problems.
Additionally, our test coverage has improved thanks to AI-driven test generation, which allows us to write tests earlier in the development cycle. This has led to fewer bugs in production and faster release times. By analyzing these metrics, we can see that AI tools have led to faster, higher-quality development, with less time spent on manual tasks and more focus on strategic problem-solving.
VMblog: Does your product consider relying on AI agents? What new potential features / abilities could this open for the platform?
Cole: Yes, we are actively building our new platform around an agentic model, with AI agents at its core. This shift is designed to automate complex workflows, offering users a more intuitive and seamless experience. In the near term, we’re focusing on features like document format translation, OCR (Optical Character Recognition), and intelligent data extraction. These capabilities will allow users to convert documents between formats, extract valuable insights from unstructured data, and automate repetitive tasks, ultimately saving time and enhancing productivity. Looking further ahead, we envision a world where agent-based computing makes workflows nearly invisible. Users will interact with the platform via simple, natural interfaces (voice, text, etc.), while AI agents handle tasks like scheduling, data analysis, or report generation autonomously. This could revolutionize how users interact with the platform, making it both more powerful and frictionless in the long term.
VMblog: What are some risk factors to consider when your team tightly works with AI for product development?
Cole: When working closely with AI in product development, the primary concern is data privacy and security. AI tools often require access to large datasets to train and improve models, and these datasets may contain sensitive customer information. Ensuring robust data governance is crucial for preventing unauthorized access and ensuring compliance with regulations like GDPR. We’ve implemented strict data protection protocols, including encryption, access controls, and regular security audits to mitigate risks. Another risk is the potential for over-reliance on AI, especially in decision-making. While AI tools are powerful, they should not replace human judgment in critical development decisions. To mitigate this, we ensure that AI acts as a support tool rather than a decision-maker, and human oversight remains integral in the development process. Striking the right balance between automation and human control is key to maintaining quality and security in our AI-powered development workflows.
##
