The Appeal of In-House AI Development

However, the narrative isn’t simply one of displacement. Some analysts suggest that the shift could go further, with organizations opting to cancel SaaS subscriptions altogether as they build their own AI-powered alternatives. While major SaaS platforms like Salesforce and Slack aren’t immediately threatened, companies leveraging AI coding assistants or developing proprietary AI agents are beginning to construct smaller, internally-connected enterprise applications. This represents a move towards greater control and customization.

Factory, a provider of AI coding agents, exemplifies this trend. According to Eno Reyes, the company’s Chief Technology Officer and co-founder, Factory is “eating its own dog food,” developing internal enterprise applications using its own technology. “We started seriously considering whether we could build software ourselves instead of buying it or subscribing to it around six to eight months ago,” Reyes explains. The company’s customer support workflow and legal tools were created internally with AI agents, and a third-party analytics application was replaced with an in-house version. “We’ve started building a lot of things internally that we historically would have purchased,” Reyes states, adding that many of their internal systems are now powered by code generated by AI agents.

This pattern is extending to Factory’s clients. Reyes observes that many are now creating tools that previously would have been sourced as small utilities or micro-SaaS products. “With an agent, someone can simply say, ‘Build me a dashboard that shows engineering velocity’ or ‘Connect this dataset to that one and visualize it.’ Instead of going through the acquisition process, the tool is simply created.”

The advantages of this approach are clear: flexibility and speed. “If you want something very specific, an agent can generate it directly from your own data, systems, and workflows,” Reyes emphasizes. “That’s why things like internal dashboards, analytics tools, or small workflow applications are often easier to create than to buy these days.”

The Costs and Complexities of DIY Software

However, building and maintaining software in-house isn’t without its drawbacks. Reyes cautions that the costs associated with developing and supporting internal applications are significant. While AI agents can generate code, a comprehensive SaaS product represents years of dedicated effort from a large team. “Even when we benchmarked agents replicating SaaS products function by function, they could do it, but it took a long time and was expensive. And when it was finished, we still didn’t have a team of hundreds of people maintaining the system,” he notes.

Another potential issue is the scope of the software being created. While AI-powered tools excel at producing smaller applications, complex enterprise systems require robust infrastructure to manage and maintain the code over time. Reyes doesn’t foresee AI-driven coding replacing established SaaS platforms like Slack, which benefit from strong network effects, or CRM systems like Salesforce, which serve as central information hubs.

Instead, he predicts that AI assistants and agents will compete with the applications that surround these core systems – tools that connect other products, visualize internal data, or provide small workflow utilities. These are the areas where on-demand generation is most feasible.

Adam Arellano, Field CTO at Harness, a provider of AI-powered development tools, also observes a growing trend towards in-house enterprise software development, but warns of the risks. “This is happening frequently, with some extreme cases where a senior executive has ordered, ‘No new software will be purchased or personnel hired; do it with AI.’ There are also more reasonable approaches where a company has created point solutions for very specific problems and achieved short-term success, but sometimes those solutions falter over time.”

Arellano highlights the immediate gratification of creating a tool tailored to a specific need as a key benefit. However, he also points to the challenge of maintaining that software and integrating it with other applications. “This isn’t unique to vibe coding tools; it’s always been the hard part of point solutions within an enterprise. But vibe coding exacerbates the problem because the speed at which these tools can be produced is much faster than a company’s ability to integrate the results, understand how they work, and maintain the connections.”

He believes that, in the near term, AI-driven coding is unlikely to deliver significant cost or time savings for critical internal applications unless the process is carefully managed. He also references recent AWS outages linked to AI-generated code.

However, Arellano acknowledges that improvements in AI-powered coding assistants will eventually make it easier for companies to develop their own enterprise software. “It will take time, and like any new technology, the path to perfection will be littered with the wreckage of tools that were ‘almost good enough.’”

A Phase of “Seduction” and the Looming Threat of Tech Debt

Not all IT leaders are optimistic. Geoff Burke, Senior Technology Advisor at ransomware defense provider Object First, describes the current enthusiasm for AI-driven development as a “phase of seduction.” “It seems like a brilliant partner at first,” he says, “but if given too much autonomy, it introduces inaccuracies and complexity, and bypasses security protocols, leading to twice the work to fix later.”

Burke insists that AI-assisted development must operate under strict access controls, rigorous peer reviews, robust testing, and isolation from sensitive information and production environments.

He warns against rushing to incorporate AI simply to appear modern. “While it may be appropriate in some parts of the stack, in core development workflows and repositories, CISOs shouldn’t follow trends based on experimental AI making critical decisions about code and data integrity.”

Blake Crawford, co-founder and CTO of IT consultancy Fusion Collective, believes that collaborative programming with strict controls can be effective, but warns of chaos if employees outside the IT team discreetly create their own solutions using AI coding assistants. He emphasizes the potential for devastating technical debt when employees are free to create their own enterprise applications without oversight.

“Most experienced IT professionals understand the strengths and weaknesses of AI-generated software, but an accountant creating complementary applications for their SAP workflow may not,” Crawford explains. He acknowledges using AI coding assistants in his daily practice, drawing on over 25 years of tech experience. “I understand where AI coding assistants excel, and more importantly, how to define what ‘good’ looks like in software development. That allows me to quickly identify problems and avoid adding to technical debt.”

Crawford notes that AI assistants are proliferating in companies that aren’t primarily focused on IT products, often leaving employees and managers unsure how to best utilize them. “With automated programming, the company owns what’s created, including the problems it generates. A company doesn’t function well with a proliferation of applications, many of which will be misused and grow beyond their scope. The result will be a complete mess, from support to integration.”

Crawford sees the temptation to develop in-house enterprise applications growing as AI coding assistants improve, but urges caution. “There will be a serious reckoning when the bill comes due for poor architectures and accumulated technical debt. If companies aren’t careful, executives will face problems that will take years, if not decades, to resolve.”

What are your thoughts on the future of AI-driven development? Do you see your organization embracing this trend, or are the risks too significant? Share your perspective in the comments below.

Frequently Asked Questions About AI and Software Development

• Q: Will AI coding assistants completely replace traditional software developers?

  A: While AI coding assistants are becoming increasingly powerful, they are unlikely to completely replace human developers. They are best viewed as tools to augment and accelerate the development process, rather than as a full replacement for skilled programmers.

• Q: What are the biggest risks associated with building software in-house using AI?

  A: The primary risks include accumulating technical debt, integration challenges, security vulnerabilities, and the ongoing costs of maintenance and support. Without proper governance and expertise, these risks can quickly outweigh the benefits.

• Q: How can organizations mitigate the risks of AI-driven development?

  A: Implementing strict access controls, conducting rigorous code reviews, prioritizing robust testing, and isolating sensitive data are crucial steps. A well-defined governance framework and a phased implementation approach are also essential.

• Q: Is “vibe coding” a viable alternative to purchasing SaaS solutions?

  A: For small, highly specific applications, “vibe coding” can be a cost-effective and efficient alternative. However, for complex enterprise systems, the long-term costs and challenges of maintenance and integration often make SaaS solutions a more practical choice.

• Q: What impact will AI have on the future of the SaaS industry?

  A: AI is likely to reshape the SaaS landscape, driving increased competition and forcing providers to innovate. We can expect to see more SaaS offerings incorporating AI-powered features and a greater emphasis on customization and integration.

• Q: How can companies ensure the security of AI-generated code?

  A: Security must be a primary consideration from the outset. This includes implementing secure coding practices, conducting thorough vulnerability assessments, and regularly updating AI models to address emerging threats.