I make my own mayonnaise. And I’m not being precious about it—my homemade mayo is delicious. Fresh eggs, good olive oil, a little Dijon, a squeeze of lemon. It takes about three minutes with an immersion blender, and I can tweak it exactly how I like it.

So why is there a jar of Hellmann’s in my fridge right now?

Because it’s Tuesday. Because my kid needs a sandwich. Because the jar is ready at all times, and it works, and I have other things to do.

I think about this every time someone tells me they’re going to “vibe code” their way out of buying enterprise software.

Let’s be clear, vibe coding is genuinely powerful

I’m not here to dunk on the movement. I’ve watched people with no formal engineering background stand up functional apps in an afternoon that would have taken a dev team weeks a few years ago. The barriers to creating software have never been lower, and that’s a legitimately great thing.

If you need an internal dashboard, a custom screenshot tool, a quick integration between two APIs—go forth and vibe. Seriously. You’ll probably build something that fits your needs better than anything off the shelf, and you’ll do it fast.

But somewhere in the excitement, a question started circulating that deserves a more honest answer than it’s been getting: “If I can build software this easily, why would I buy from a vendor?”

It’s a fair question. Here’s my answer.

Nobody should be vibe coding a…

Contact center. You’re not just routing calls. You’re handling skills-based routing across channels, real-time sentiment analysis, workforce management, compliance recording, SLA enforcement, and seamless escalation paths between bots and humans—all while maintaining context across every touchpoint. Good luck prompting that into existence.

Order management system. Distributed inventory, split shipments, return authorization workflows, tax calculation across jurisdictions, fraud detection, and integration with a dozen logistics partners who all speak different EDI dialects. Your vibe-coded prototype handles the happy path beautifully. The happy path is 20% of the job.

Revenue recognition engine. ASC 606 alone should end this conversation. But sure, go ahead and explain to your auditors that the AI wrote your rev rec logic.

And then there’s the one I know really well…

I spend my days in consumer goods. Trade promotion management (TPM) is one of the most critical—and most notoriously complex—functions in the Consumer Package Goods (CPG) industry. Manufacturers spend upward of 20% of gross revenue on trade promotions, making it the second-largest line item on the P&L after cost of goods sold. And for most companies, between 40–70% of those promotions lose money.

Now, if you describe TPM to an LLM—“build me a system to plan, execute, and analyze trade promotions”—you’ll get something back. It’ll have a UI. It might even look pretty good. It will handle exactly none of the things that make TPM actually hard:

Fund management with accruals, commitments, and actuals tracked across overlapping fiscal calendars, retailer calendars, and promotional windows—where a single fund might cascade through hierarchies of customers, products, and geographies with different rules at each level.

Deduction matching that reconciles retailer claims against planned promotions when the retailer’s description of what they ran bears little resemblance to what was actually agreed upon. This is an entire discipline unto itself. Companies employ teams of people whose sole job is closing deductions.

Post-event analysis that separates baseline volume from incremental lift, accounts for cannibalization across the portfolio, and feeds optimization models that inform the next planning cycle—all calibrated to each retailer’s unique data-sharing cadence and format.

Regulatory and contractual compliance that varies not just by country, but by channel, by retailer, and sometimes by product category within the same retailer.

None of this is in the code. All of it is in the domain knowledge. And that’s the part no one talks about when they’re excited about vibe coding: the complexity of enterprise systems doesn’t live in the technology. It lives in the decades of institutional expertise encoded into the primitives, logic, workflows,  exception handling, and the hard-won implementation lessons about what actually works when you’re managing a billion-dollar trade spend.

The day two problem

Let’s say you do build it. Congratulations—now you get to maintain it.

You need to write documentation explaining what a “promotion” is in your system—because your definition is different from every other company’s definition, and the person who built this is about to go on paternity leave. You need markdown files explaining the data model, the business rules, the edge cases. You need to handle the integration that breaks when your ERP vendor pushes an update. You need to onboard the next hire and explain why the deduction matching logic has seventeen conditional branches. You need to keep up with regulatory changes, with new retailer requirements, with evolving data formats.

Vendors aren’t just selling you software. They’re selling you a team of people who wake up every morning thinking about these problems so you don’t have to. They’re selling you a community of customers who have collectively surfaced and solved edge cases you haven’t even encountered yet. They’re selling you a roadmap informed by the entire market, not just your corner of it.

The real question isn’t “can I build it?”

The question was never about capability. It’s about allocation.

Where do you want your best people spending their time (and in the AI era, tokens)? Maintaining plumbing—or activating against the insights that plumbing delivers? Rebuilding objects that already exist in mature platforms—or building the custom workflows that make those platforms sing for your specific business?

Because here’s the thing: vibe coding is incredible for that last mile. It’s incredible for the custom workflow that sits on top of your platform. For the integration that connects your CRM to that internal tool nobody else uses. For the agent that automates a specific, well-defined task that’s unique to your business.

Salesforce is leaning into this hard. Agentforce lets you build custom Apex actions to extend agent capabilities—your team or your SI partners can create bespoke logic that sits on top of the platform’s foundation. You’re not rebuilding the house; you’re customizing the rooms. That’s where the speed and flexibility of AI-assisted development truly shines: not replacing the platform, but extending it in ways that would have required a six-month project two years ago.

The smart play isn’t “build everything” or “buy everything.” It’s buy the platform, own the edge.

Back to the Mayo

Make your own mayonnaise on a Saturday afternoon when you’re feeling creative and you’ve got the good eggs. It’ll be better than anything you can buy. Enjoy it.

But when it’s Tuesday and you need to feed the family, buy the Hellmann’s. It’s been perfected over a hundred years by people who think about emulsification for a living. It’s consistent, it scales, and it frees you up to focus on the parts of dinner that actually need your personal touch.

Know the difference. Your trade spend—and your Tuesday nights—will thank you.