The landscape of software development is undergoing a seismic shift, moving beyond the realm of dedicated engineers to encompass the C-suite and other senior executives. This emerging trend, dubbed "vibe coding," sees business leaders leveraging artificial intelligence (AI) coding assistants not for traditional programming, but to rapidly prototype and deploy a diverse array of agents and productivity applications. This phenomenon is reshaping how businesses operate, accelerating innovation, and challenging conventional notions of software creation.
What began as a tool for developers has rapidly expanded its reach. Executives, often driven by a mix of impatience with lengthy IT queues and genuine curiosity about AI’s capabilities, are now at the forefront of this movement. Their endeavors range from automating simple workflows to establishing full-fledged production systems capable of serving hundreds of users. Tools like Claude and Cursor, alongside the increasingly integrated AI features within existing enterprise platforms, are becoming the instruments of choice for this new wave of AI-empowered business builders. While the enthusiasm is palpable, the results, as we will explore, are as varied as the motivations behind them.
From Zero to 140,000 Lines: A CEO’s AI-Driven Creation
Moshe Bar, CEO and co-founder of Codenotary, provides a striking example of this executive-driven AI adoption. Bar, who explicitly states he is not a programmer, embarked on a project in March 2025 to build a bulletin board system (BBS) specifically for users of IBM 3270 mainframe terminals. His role was to define the core database structure – encompassing tables for users, messages, topics, posts, and chat functionalities – and then delegate the intricate task of code generation to Claude. The outcome is a remarkably complex system, now boasting 140,000 lines of code, with Bar having personally edited a mere 10 lines.
"A project of this size, with this reliability, with this usefulness in such a short time would have been unthinkable just two years ago," Bar remarked to The New Stack. The system, accessible at moshix.tech:3270, currently supports 500 users, facilitates thousands of active discussions, and has maintained an uptime of 100% since its inception. Bar also employs Claude for bi-weekly security audits, ensuring the system’s robustness. Demonstrating a commitment to open innovation, he has open-sourced the code, enabling others to leverage and adapt it for their own BBS instances.
The 10 lines Bar personally modified were solely for menu text adjustments, made necessary during transatlantic flights due to rate-limiting timeouts encountered with Cursor. Since transitioning fully to Claude, his direct code interaction has been virtually non-existent. The primary technical hurdle has been the 3270 protocol itself, an obscure and highly specific IBM standard dating back to the early 1970s. Claude occasionally exhibits minor issues, such as reverting to Unicode or misplacing cursor positions, requiring hundreds of corrections from Bar. He colorfully describes the AI as "a kindergartner with a doctorate." Despite these occasional challenges, the application operates efficiently, consuming a mere 23 megabytes of memory and having experienced no security incidents in over a year. Bar estimates that a conventional development approach for such a system would have incurred costs between $1.2 million and $2 million, requiring three to four senior developers at salaries ranging from $400,000 to $500,000 each.
The Dual Build: A CEO’s Experiment in AI Efficacy
Woodson Martin, CEO of OutSystems, adopted a more structured approach to his own "vibe coding" experiment. He developed a personal mobile application wrapper for his team’s existing MCP services. Intriguingly, he undertook this project twice in parallel: once using OutSystems’ proprietary AI coding tool, Mentor, and again with Claude, both connecting to the identical backend.
"I was tired of explaining it to somebody who was supposed to build it for me," Martin explained. "I was just like, ‘I’ll do this myself.’" His creation serves as a personal chief-of-staff system, consolidating customer account intelligence—including buying signals, website activity, and internal data—into a pre-meeting briefing accessible on his mobile device. This new system effectively replaces a previous workflow that involved a 45-minute PowerPoint presentation and multiple preparatory meetings with his sales team.
Martin also utilizes the V2MOM (Vision, Values, Methods, Obstacles, Measures) goal-setting framework, a methodology he adopted from his time at Salesforce and which his team has since productized as an MCP service. He employs it to identify alignment gaps between individual and corporate goals during executive one-on-one meetings. An agentic layer, built around this framework, is now instrumental in cascading organizational change across the company.
At Codenotary, Bar has implemented a company-wide mandate for LLM-only development, a decision stemming directly from his AI-assisted coding experiences. Approximately three to four months prior to the report, he declared that all future development would exclusively utilize large language models. His rationale is rooted in the accelerated pace of technological evolution: "Cloud computing compressed time-to-market from three years to one; LLMs compressed it again to three months." He firmly believes that "If you cannot come to market with a new application from scratch within three months, you’re probably already missing the market." Despite his success in "vibe coding" the BBS, Bar humbly admits to not feeling like a programmer, stating, "No, I don’t. Sadly, I don’t, because when I look at the code, some of the things it does, I have no idea, no idea."
Democratizing Development: Smaller Scale, Significant Impact
Beyond these large-scale projects, executives are engaging in "vibe coding" on a smaller scale, yet with a similar intent to enhance personal and team productivity. Wade Foster, co-founder and CEO of Zapier, described building a personal AI chief of staff using the Zapier SDK within Cursor. He noted on LinkedIn, "The SDK handles auth for every tool I use at work. The coding agent handles everything else. I just describe what I want."
Jessica Stefanowicz, external communications manager at Anaconda, successfully "vibe coded" an awards tracking dashboard in Claude, marking her first project of this nature. This tool autonomously updates weekly, monitors application deadlines and submission requirements, and proactively recommends award programs that align with specific business updates and products. This automation has significantly reduced her manual workload, saving approximately an hour per week.
David Slater, chief marketing officer at Front, a customer service and operations platform, has developed a two-agent system using Claude to manage between 50 and 70 Objectives and Key Results (OKRs) over a 90-day cycle. One agent is designed to prepare for bi-weekly check-ins by flagging areas requiring attention and generating relevant questions for owners. The second agent is responsible for producing tailored status reports for three distinct audiences—product, sales, and the executive team—following review completion. Notably, Slater achieved this without writing a single line of traditional code. "I’m not a technical user. I didn’t build these agents by writing code," Slater told The New Stack. "But this system has fundamentally changed how I run a core part of our department’s operating rhythm."
Skepticism and Safeguards: Navigating the Risks of "Vibe Coding"
While the trend of executive-led AI development is gaining momentum, analysts offer a more nuanced perspective, highlighting potential risks and limitations. Andrew Cornwall, an analyst at Forrester, acknowledges that non-technical executives are indeed building significant web applications. However, he also catalogs the inherent risks: "vibe-coded" applications often lack the robust security hardening required for enterprise environments, may not meet auditor compliance standards, and can end up being handed over to CIOs or CTOs without adequate budget or maintenance plans. In the most concerning scenarios, executives might utilize unapproved AI providers, inadvertently leading to corporate data breaches.
Cornwall offers a pragmatic view: "If vibe coders and their users understand the limitations of their apps, they’re not riskier than the spreadsheets they built in the past." The more significant challenge, he posits, lies in the potential for a polished-looking application to obscure whether it is professionally supported or merely a collection of AI-generated prompts.
His Forrester colleague, Ken Parmelee, adds two critical skeptical notes. Firstly, he observes that executives are rarely building in complete isolation; an IT professional is often involved, though the executive may claim sole credit in internal communications. Secondly, and more fundamentally, Parmelee contends that most non-developers encounter a developmental wall. "Non developers are not systems thinkers," he stated. The moment a project necessitates a database, persistent memory, or complex backend integration, the majority of these executives find themselves unable to proceed independently. He questions, "Can they do very simple applications or flows, but many times that doesn’t equate to real value."
Broader Implications and the Future of Software Development
Despite these concerns, Holger Mueller, an analyst at Constellation Research, identifies a significant, often overlooked upside to this trend. When executives build their own automation solutions, they do not consume developer budgets or commandeer engineering capacity. "Projects stay on course" for the rest of the enterprise, Mueller noted.
Brad Shimmin, an analyst at the Futurum Group, observes a rapid outward expansion of agentic coding, or "vibe coding." Their recent Decision Maker survey of 818 data professionals revealed that 75% are already dedicating more time to business concerns rather than traditional coding pipelines, thanks to generative and agentic AI tools.
Shimmin further indicated that this acceleration is not confined to IT professionals: "Anecdotally, we are seeing this same acceleration and expansion among non-IT professionals, especially executives. No, these folks are not writing software in the traditional sense. On one end of the spectrum, they’re automating a wide range of tasks to free up time and energy. And on the other end, they’re using these tools to drive change and explore new opportunities. Instead of calling a meeting to discuss a new idea (e.g., a new product or a revised workflow), executives are building their own working prototypes to stress-test their ideas and, if successful, accelerate time-to-market by handing over these early assets to IT for refinement and implementation."
A New Baseline or an Isolated Phenomenon?
Moshe Bar remains clear-eyed about his own position within this evolving landscape, acknowledging, "I think I’m a little bit of an outlier due to the scope of the application." He observes that most "vibe coders" are addressing specific, well-defined needs with smaller applications, making entire systems that serve hundreds of users a rarity.
However, the baseline for what is technically achievable continues to rise. Bar reflects on a conversation from a few years prior, where he confidently assured a neighbor that the demand for developers would remain consistently high. "I’m starting to think maybe that was a mistake," he confessed, "because I start to think maybe we’re slowly coming to an end of this oversized demand for developers."
Gene Kim, author of "Vibe Coding: Building Production-Grade Software," frames this transformative moment expansively, stating, "We’re witnessing a software revolution that may make the 1990s internet boom look like a warm-up act." In a previous interview with The New Stack, Kim posited that AI coding represents a transformation "10 to 100 times bigger than DevOps," and he is placing his professional reputation on the potential of "vibe coding," where individuals can effectively "speak computer programming." The executives who are "vibe coding" in their off-hours may well represent the vanguard of this profound technological and operational shift.