The first post in this series argued against building your own Salesforce or HubSpot: Don't Build What You Can Buy. The second covered why even a better product loses to the tool a team already knows how to use: Why Incumbents Still Win. Put those two together and you get what I actually recommend: keep the platform you have and let AI bridge the gaps.

This isn't about replacing the tools you already have. It's about using AI to make them work the way your business actually runs.

The platform is whatever standard tool you already pay for. HubSpot, Salesforce, Zendesk, your ERP. It handles auth, storage, permissions, search, audit logs, compliance, and reporting. All the stuff your team relies on without thinking about it. That's the whole reason this approach works. You inherit all of that instead of rebuilding it from scratch.

The custom layer is AI closing the gap between what the platform does and what you need. It could be an external AI tool that connect to the platform's API. Sometimes it's custom code you build, which AI now makes cheap enough to be worthwhile. The point is you're using AI to bridge the gap, not automatically jumping to custom software.

Most companies need maybe 20% custom functionality on top of 80% standard functionality. Before AI, getting that 20% meant either accepting a bad fit or spending months building custom software. Now AI gives you multiple ways to get that 20% without the old tradeoffs.

Your sales workflow doesn't match the standard CRM process? Maybe the vendor's new AI features handle it. Maybe an external AI tool connects to the API and automates the difference. Maybe you do need to build something custom, but now it takes days instead of months. The answer isn't always "build custom software." The answer is "use AI to bridge the gap."

I've automated complex approval workflows using AI agents that already existed in a platform. I've connected external AI tools to a CRM to handle customer interactions in ways the CRM never anticipated. No code written.

And yes, I've built custom integrations when nothing else worked. But I only built what I needed to build.

The economics changed because AI gives you options. You're not stuck choosing between a bad fit and an expensive custom build. You can try the platform's AI features first. If those don't work (they rarely do thus far), you can try external AI tools. If those don't work, then you build something custom. But now that custom build takes a fraction of the time it used to.

Your team doesn't retrain on fundamentals. The spreadsheet macros still work, the dashboard configurations stay put, the keyboard shortcuts don't move. From their perspective, the platform just got better at handling their specific needs. Whether that happened through the vendor's AI, an external tool, or custom code doesn't matter to them.

The vendor relationship stays intact. They handle support, compliance, security updates, and feature development. Your team doesn't inherit those responsibilities. You focus on using AI to handle what makes your business different, not on maintaining authentication systems and database backups.

This approach sidesteps almost every problem that kills internal software projects. There's no data migration because the data never moves. Nobody needs training on a new system. The only thing that changes is the platform now handles work it couldn't handle before.

When you do need to build custom code, you're building the minimum. A script that moves data between systems. An automation that triggers based on your business rules. An interface that presents information exactly how your team needs it. You're not building a platform. You're building the 20% that makes you different.

The best use cases are boring. Custom approval workflows that route based on your specific business rules. Reports that combine data from multiple systems in ways the vendors never imagined. Automations that handle the repetitive work your team does every day. Notifications that trigger on conditions specific to your business.

None of this is free. When you build custom code, you own it. It needs maintenance, monitoring, and updates when APIs change. But you're only owning the pieces you absolutely had to build. Everything else either comes from the vendor or from AI tools that someone else maintains.

The custom layer also inherits the platform's constraints. Rate limits, API restrictions, vendor lock-in. These are real. But they're more manageable than the constraints of building and maintaining a complete replacement.

This approach assumes you already have a platform that handles most of what you need. If existing platforms miss 60% of your requirements, you probably do need to build from scratch. It also assumes your business processes are stable enough that constant rewrites won't eat up the savings.

Most teams compare the cost of building versus buying. But that's the wrong comparison when AI gives you so many options in between. The right comparison is the cost of the platform plus whatever AI solution bridges the gap versus the cost of building everything from scratch.

You get the stability and feature development of a mature vendor plus AI handling what makes your business unique. The platform handles the foundation. AI bridges the gap. You build custom code only when you have to.

Buy the platform. Let AI bridge the gap. Build only what you must.