Vibe coding has quickly emerged as one of the most talked-about applications of AI in enterprise software. In simple terms, it refers to using AI tools to generate code from natural language prompts. On paper, it sounds transformative – describe what you want, and the AI builds it for you. But as with most AI promises, the reality is a lot more complicated.

We recently published an article titled “I Love Vibe Coding, But It’s Dangerous and You Probably Shouldn’t Do It”, which captures much of the current sentiment surrounding the trend. While the potential is exciting, it’s difficult to ignore the risks that come with AI-generated code, particularly around security, governance, and oversight.

That hesitation appears to be reflected among a lot of Salesforce Admins at the moment. According to the 2026 SF Ben Salesforce Admin Survey, AI adoption overall is relatively high – 44% of respondents said that they use AI daily or regularly, while a further 41% use it occasionally. However, when it comes to vibe coding specifically, adoption is far lower. While it is, of course, a tool more closely associated with developers, 71.3% of admin respondents said they are not using vibe coding tools at all, which feels pretty high.

Realistically, there may be several reasons for this, all tied to the current complex environments that modern Salesforce Admins have to work and compete in at the moment. So to explore this further, we spoke with Salesforce professionals about why admins remain cautious around vibe coding, what the current barriers are, and what would need to change before adoption becomes mainstream – or at least improved.

SF Admins Are Overloaded

One of the key reasons vibe coding adoption remains low among Salesforce Admins may actually simply come down to bad timing. AI tools are arriving at a moment when admins already feel stretched beyond the traditional boundaries of the role.

Our Admin Survey data shows that over half of respondents (58.6%) said that Salesforce is becoming more complex, while 53.1% agreed that too much is now expected of Salesforce Admins.

Meanwhile, tech debt was identified as the single biggest challenge facing teams today, adding yet another layer of difficulty for professionals already trying to keep complicated orgs running well.

On top of this, many admins lack clear support. 42.86% of respondents work in admin-only teams, while 20% are operating as solo admins entirely. That means many are simultaneously acting as admins, analysts, release managers, security owners, and sometimes even accidental architects.

With this in mind, vibe coding can feel less like a productivity breakthrough and more like another responsibility being added to an already overloaded plate.

Charlie McGregor, Salesforce Solution Architect and Co-Founder at Catalyst, believes this overload is one of the biggest reasons admins are hesitant to embrace AI-generated code.

“Everything’s being thrown at them,” he explained. “There’s all these different tools, all these different bits and pieces around that are making it very hard to keep up.”

He also pointed to the growing fatigue surrounding AI adoption in general, particularly as businesses push employees to adapt quickly while simultaneously raising expectations around productivity.

“There’s a real push for it, and we’ve seen a lot of companies using AI as an excuse for layoffs,” he said. “I think a lot of people are staying away from it from a self-preservation perspective as well.”

This is especially relevant in Salesforce, where the admin role has evolved massively in the last few years. The role now increasingly asks for knowledge around automation architecture, integrations, DevOps processes, and so on.

Charlie argued that modern admins can no longer afford to only understand part of the platform, saying: “You need to understand the whole platform. You need to understand order of execution, handler classes, Apex triggers, and best practices. Otherwise, if you’re using AI to generate code, how are you validating that it’s actually doing what you want it to do?”

This is then where a “quick wins versus long-term risk” debate emerges. AI can absolutely accelerate smaller tasks or lightweight development work. But in large, customized Salesforce environments that carry years of tech debt, introducing AI-generated code without the right governance can create entirely new risks.

Benjamin Bates, Salesforce Architect at Flight Centre Travel Group, believes many admins simply don’t see vibe coding as naturally fitting within their existing responsibilities.

“If I call myself a Salesforce Admin, I’m probably not touching code or expected to,” he explained. “My domain is config and metadata rather than classes and custom components.

“If I’m in Claude Code, in a CLI, working in Linux terminals, that’s where I feel comfortable. Admins are probably not in that world.”

In essence, while AI may make it easier to generate code, it doesn’t remove the need for governing and validating when it gets created. For admins who are already stretched, that additional responsibility may just outweigh any potential productivity gain.

Does Vibe Coding Assume Technical Skills?

Another big question surrounding vibe coding for admins in Salesforce is whether it quietly assumes a level of technical knowledge that many admins just don’t have, or were never expected to have in the first place.

The Salesforce ecosystem has historically been built around the “clicks, not code” philosophy, with admins primarily operating in declarative environments like Flow, validation rules, and page layouts. According to our Admin Survey data, this still remains true today – around 69% of admins said they work mainly or fully in declarative tools, while only 8% said they work primarily in programmatic environments.

Charlie argued that AI-generated development still requires a solid understanding of Salesforce architecture and coding fundamentals.

“You don’t need to be a developer, but you need to understand structure,” he explained. “If you’re getting AI to write something for you, how are you validating that it’s actually doing what you want it to do? If you can’t validate Apex, you probably shouldn’t use AI to generate it.”

This may sound harsh at first, but what the AI delivers is not always correct, secure, or maintainable, and must be well understood and monitored closely. Charlie describes, for example, multiple occasions where AI tools confidently produced incorrect answers or hallucinated entirely.

“I’ve done projects recently where what AI is telling me is just straight wrong,” he said. “And if you don’t understand what it’s giving you, where are your safeguards?”

Benjamin echoed the same concern, particularly around blindly trusting AI-generated outputs without proper platform knowledge.

He said: “The dangerous over-reliance comes when AI becomes your first port of call for knowledge,” he explained. “If it’s not backed up by a pretty good understanding of the platform, you’re probably going to end up with egg on your face.”

Importantly, neither argued that admins should avoid AI altogether – in fact, both use AI in their own work. But the difference is that they view AI as an accelerator for existing expertise rather than a replacement for it.

Benjamin described how AI can significantly speed up prototyping and development workflows for technically confident users, particularly as the models become better at understanding Salesforce metadata and deployment structures.

“The ability for these models to understand Salesforce XML and metadata has improved massively,” he said. “If you give it the right prompts and the right context, it can do a lot more than it could even six months ago.”

But even with those improvements, both architects emphasized that AI still struggles with the broader context of enterprise Salesforce environments. Large orgs often contain years of legacy architecture decisions, technical debt, and undocumented business logic that AI simply cannot fully understand from metadata alone.

That creates a major risk if organizations begin assuming AI can safely replace technical expertise.

Trust and Governance Are Massive Blockers

Overloaded workloads and technical barriers explain part of the hesitation around vibe coding, but trust and governance may be the bigger issue hiding underneath it all.

Across the Salesforce ecosystem, trust has consistently remained a huge barrier to broader AI adoption, namely Agentforce. And when AI moves from generating simple prompts to producing code, the stakes become a lot higher.

According to our survey, 58% of respondents said they have serious concerns around AI security and governance. At the same time, security management remains one of the lowest-confidence skill areas among admins overall, highlighting a worrying disconnect between AI performance and how ready businesses are for it.

And in many cases, guidance on governance isn’t fully mature yet. For example, many would say that awareness around concepts like Zero Trust and Salesforce’s Shared Responsibility Model remains relatively limited across the ecosystem. So the idea to then introduce AI-generated code into production environments understandably raises alarm bells.

For Charlie, the real issue isn’t the tooling itself, but whether businesses will know how to use it maturely and safely, stating: “Ninety percent of businesses out there aren’t ready for AI.”

Charlie repeatedly returned to the idea that AI is being treated as a “silver bullet”, especially by businesses looking to move faster, reduce spend, or cut headcount – without fully understanding the risks involved.

“People have been treating AI like this silver bullet,” he explained. “But it’s not going to make your entire dev team obsolete. It may streamline work, but it’s not removing the need for governance, validation, or oversight.

“If you’re getting AI to do everything – taking a ticket, pasting it into AI, and deploying whatever comes back – that’s dangerous. Anything you’re getting AI to do, you should know how to do it yourself.”

He also echoed the importance of operational discipline and how important it is to have proper processes. In fact, he mentioned that he intentionally disables deployment permissions within his own AI tooling to prevent accidental production changes.

Benjamin raised similar concerns around this discipline, particularly about how confidently AI systems can present incorrect and incomplete information.

“The systems are very opinionated,” he explained. “They’ll confidently tell you Salesforce can solve a problem a certain way, but unless you already understand the platform, you may not realize the answer falls apart once you actually try to implement it.

Still, despite the caution, there was optimism about where the technology is heading. Benjamin believes Salesforce Administration will become increasingly conversational and AI-assisted over time, while Salesforce itself already appears to be moving toward more guided, governed experiences through tools like Agentforce for Setup. 

That distinction may prove critical, because while vibe coding may eventually become commonplace, many admins today feel the technology is advancing faster than the governance frameworks needed to support it safely.

Final Thoughts

So while Salesforce Admins are embracing AI in a variety of ways, vibe coding may still feel like a risk too far for many at the moment.

We’re already seeing examples of vibe coding causing issues when deployed without proper oversight, and many admins likely already have too much on their plate to worry about becoming the next headline because of a poorly governed AI slip-up.

That doesn’t mean the ecosystem will stay this way forever, though. Like AI adoption more broadly, vibe coding will likely mature over time as tooling, governance, and best practices improve. But for now, many admins remain focused on tackling other major challenges across security, complexity, technical debt, and increasing role expectations – all themes we uncovered through the SF Ben Admin Survey.

Keep an eye on our upcoming coverage as we continue exploring the biggest pressure points facing Salesforce Admins today!

The Authors