Are you looking for smarter insights delivered straight to your inbox? Subscribe to our weekly newsletters for essential updates on enterprise AI, data, and security.
The Experimentation with Vibe Coding at Intuit Mailchimp
In the past year, Intuit Mailchimp has been exploring the innovative concept of vibe coding. As a provider of email marketing and automation services, Mailchimp is part of the larger Intuit organization, which has been steadily advancing its generative AI initiatives through the development of its GenOS and agentic AI capabilities across various business units. Despite having its own AI solutions, Mailchimp recognized the necessity of utilizing vibe coding tools for specific challenges.
The Need for Rapid Prototyping
The journey into vibe coding began under tight deadlines. Mailchimp needed to quickly demonstrate a complex customer workflow to stakeholders, but traditional design tools like Figma were unable to deliver the required working prototype in time. Some engineers at Mailchimp had already been experimenting with AI coding tools in a more informal capacity. Faced with the urgency of the situation, they decided to apply these tools to a real business challenge.
Surprising Results and Accelerated Development
Shivang Shah, Chief Architect at Intuit Mailchimp, shared insights with VentureBeat about the urgency they faced: “We actually had a very interesting situation where we needed to prototype some stuff for our stakeholders, almost on an immediate basis; it was a pretty complex workflow that we needed to prototype.” The engineers employed vibe coding tools and were astounded by the efficiency of the results. “Something like this would probably take us days to do,” Shah explained. “We were able to do it in a couple of hours, which was very, very interesting.” This initial success prompted Mailchimp to adopt AI coding tools more broadly, leading to development speeds that are now up to 40% faster. They also learned vital lessons about governance, tool selection, and the importance of human expertise—insights that can benefit other enterprises.
Shifting Dynamics in Developer Interaction with AI
Mailchimp’s experience illustrates a significant shift in how developers engage with AI. Initially, engineers utilized conversational AI tools for basic guidance and algorithm suggestions. Shah noted, “Even before vibe coding became a thing, a lot of engineers were already leveraging existing conversational AI tools to ask, ‘Is this the right algorithm for the problem I’m trying to solve?’” However, the advent of modern AI vibe coding tools has transformed this interaction from simple consultations to a more hands-on approach where the tools assist in actual coding tasks.
Embracing Multiple AI Coding Platforms
Rather than standardizing on a single platform, Mailchimp strategically adopted multiple AI coding tools, including Cursor, Windsurf, Augment, Qodo, and GitHub Copilot. This decision stemmed from their understanding of the varying benefits that different tools offer throughout the software development lifecycle. “What we realized is, depending on the life cycle of your software development, different tools give you different benefits or different expertise, almost like having an engineer working with you,” Shah explained. This approach reflects the way enterprises often deploy specialized tools tailored to different development phases, avoiding a one-size-fits-all solution that may excel in some areas while falling short in others.
Governance and Human Oversight
One of the most important lessons Mailchimp learned about vibe coding relates to governance. The company has established both policy-based and process-embedded guardrails that other organizations can adapt. Their policy framework mandates responsible AI reviews for any AI-based deployments involving customer data. Additionally, process-embedded controls ensure that human oversight remains integral to the development process. While AI may conduct initial code reviews, human approval is necessary before any code is deployed to production. “There’s always going to be a human in the loop,” Shah emphasized. “There’s always going to be a person who will have to refine it, gut check it, and ensure it’s actually solving the right problem.” This dual-layer approach addresses a common concern among enterprises seeking to leverage AI for productivity while maintaining code quality and security standards.
Limitations of AI Coding Tools
Despite their advantages, Mailchimp discovered a significant limitation of AI coding tools: while they can grasp general programming patterns, they often lack specific knowledge of the business domain. This insight highlights the ongoing need for human expertise in conjunction with AI capabilities.