With AI coding assistants rapidly advancing, some organizations are pushing the AI-aided application development concept further by engineering enterprise-grade tools to replace or extend traditional software.
While a trend toward vibe coding your own enterprise software appears to be in its infancy, some companies have reportedly replaced their traditional CRMs with homegrown applications.
In recent weeks, stock prices of traditional SaaS vendors have taken hits over worries about competition from AI. Much of concern has focused on customers spending less on SaaS as they ramp up AI deployment and on AI agents taking over some functionality of SaaS platforms.
But some experts contend that customers may also opt out of some SaaS subscriptions entirely as they roll their own replacements. While major SaaS platforms may not be in danger, smaller enterprise apps that connect to them may increasingly be deployed in house by companies using AI coding assistants or deploying home-grown agents.
IT leaders should beware, however, that significant challenges exist. Replacing road-tested enterprise software packages with vibe-coded apps is risky, with the responsibility of maintenance and support falling on the deploying organization.
Vibe coding vs. buying
Nevertheless, AI coding agent vendor Factory has been eating its own dog food by building some of its own enterprise apps, says Eno Reyes, CTO and cofounder there. Several of Factory’s customers are doing the same thing, he adds.
In the past six to eight months, the company has started asking whether it can build a software package itself instead of buying or subscribing, he says. Factory’s customer support workflow and its legal tooling were built internally with AI agents, and the company replaced a third-party analytics app with an internal build.
“Internally, we’ve started building a lot of things that historically we would have bought,” Reyes says. “A lot of our internal workflow systems are just code that the agents produced.”
He sees the same trend among customers. “A lot of the tools they used to buy are small utilities or micro-SaaS products,” he says. “With an agent, someone can just say, ‘Build me a dashboard that shows engineering velocity’ or ‘connect this dataset with this one and visualize it.’ Instead of going through procurement, the tool just gets built.”
The advantages of building your own software include flexibility and speed, Reyes adds. “If you want something very specific, an agent can generate it directly against your own data, systems, and workflows,” he says. “That’s why things like internal dashboards, analytics tools, or small workflow apps are often easier to build than buy now.”
The cost of doing it yourself
But there are also real disadvantages, with the costs of building your own internal apps and the maintenance of the software being major considerations, Reyes cautions.
“Even though agents can generate software, a full SaaS product usually exists because a large team has spent years maintaining it,” Reyes adds. “When we run our internal benchmark where agents replicate SaaS products feature by feature, the agent can do it, but it takes a long time to run and it’s expensive. And when it’s finished, you still don’t have a team of hundreds of people maintaining the system.”
Another potential problem is the scope of the software being built. “Tools like vibe-coding apps can produce smaller applications easily,” Reyes says. “But when you’re talking about complex enterprise systems, the software gets large very quickly, and you need infrastructure that can manage and maintain that codebase over time.”
With these potential downsides, Reyes doesn’t see vibe-coding AI apps replacing all established SaaS platforms anytime soon. Enterprise apps like Slack, with strong network effects, will survive, as will systems — like Salesforce’s CRM — that function as a core source of truth, he says.
Instead, AI coding assistants and agents will compete with apps that form the layer around major enterprise systems, Reyes predicts. Tools that simply connect other products, visualize internal data, or provide small workflow utilities are easier to generate on demand, he says.
Adam Arellano, field CTO at AI-powered dev tools vendor Harness, also sees some push toward organizations developing their own enterprise software, but he warns about the pitfalls.
“This is happening a lot, with some extreme cases where a C-suite leader has mandated ‘no new software purchases or headcount, do it with AI,’” he says. “There are other more reasonable approaches where a company has built point solutions for very specific problems and found certain levels of success that are useful in the short term but sometimes flounder after a while.”
The advantages of vibe coding your own enterprise software include the satisfaction of quickly building a tool for a specific need, Arellano adds. But maintaining the vibe-coded software and getting it to work with other apps can be a challenge, he says.
“Not unique to vibe-coded tools, this has always been the hard part of point solutions in the enterprise, but getting point solutions to reliably function and play nicely with larger platforms or programs of record takes work,” he says. “Vibe coding makes the problem urgent because the speed at which these tools can be produced is so much faster than enterprises can integrate their outputs, understand how they work, and maintain their connections.”
In the immediate future, vibe coding critical internal apps isn’t likely to save much money or time, unless the process is well governed, he says, noting recent outages at AWS related to AI-generated code.
However, improvements in AI coding assistants will make it easier over the long term for companies to develop their own enterprise software, Arellano says. “It will take a while and like any new tech or tool, the road to good will be littered with the broken remains of ‘almost good enough, but not quite’ tools,” he adds. “A lot of things are going to break in the meantime.”
Seduced by AI coding assistants
Other IT leaders see major risks with vibe coding enterprise software. The practice has a “seduction phase,” says Geoff Burke, senior technology advisor at ransomware defense vendor Object First.
“At first, it feels like a brilliant partner,” he explains. “But give it too much autonomy and it injects inaccuracies, complexity, and bypasses security norms, which you will spend twice as long cleaning up later.”
AI-assisted development should operate within strict access controls, diligent peer review, robust testing, and isolation from sensitive information and production environments, Burke says.
“Companies are bolting on AI to look modern, which may be fine in parts of the stack, but in core development workflows and repositories, CIOs should not chase trends that rely on experimental AI to make critical decisions about code and data integrity,” he adds.
Vibe coding with strict controls may produce good results, but if employees outside the IT team quietly create their own workarounds using AI coding assistants, chaos can ensue, adds Blake Crawford, cofounder and CTO at IT consulting firm Fusion Collective.
There’s huge potential for crippling technical debt when all employees feel free to create their own enterprise apps without supervision, he says. Most veteran IT professionals will know the strengths and weaknesses of AI-generated software, but an accounts payable clerk creating add-on apps for his SAP workflow may not, he adds.
“I use AI coding assistants in my daily practice, but I’m more than 25 years into my dedicated technology career,” he says. “I understand what AI coding assistants are good at, and even more importantly, what ‘good’ looks like in software development. That allows me to intercept problems quickly and avert any additional tech debt.”
AI assistants are showing up at many companies that don’t focus on IT products, and in many cases, employees and leaders have major questions about the best way to use them, Crawford says.
“With vibe coding, a company then owns what is made, right down to the problems it creates,” he says. “An enterprise doesn’t work well if it’s stitched together with myriad apps, many of which will be misused and grow beyond their scope, making everything from support to integration a problem.”
Crawford sees the temptation to roll your own enterprise apps growing as AI coding assistants improve, but he urges caution.
“There will be a serious retraction when the bill comes due on poor architectures and accumulated technical debt,” he says. “If companies aren’t careful, leaders will be looking at years, if not decades, worth of issues to deal with.”