I was recently talking with a friend who works with a small non-profit about the challenges they're having integrating new fundraising software. While the software works well for 80% of their needs, the last 20% is proving incredibly hard to customize—to the point where he's considering getting rid of the software entirely and replacing it with custom low-code workflows. This is the classic build vs. buy distinction which companies have faced for a long time, but I think it will actually become less salient over the next few years.

Instead, most software will offer a combination of build and buy—we'll see a return to personalized computing, with more emphasis on custom workflows and integrations. Two key technologies are driving this change:

1. WebAssembly has matured to the point where it's usable for run-of-the-mill SaaS software to provide an extension platform for users. The main open problem here is around sharing downstream packages and dependencies—only small scripts are (easily) portable today.
2. Large language models will allow users to customize the code in their software with minimal expertise. Here, the biggest remaining obstacle is providing enough constraints and context for LLMs to generate code that works well in a particular application/SDK.

Since this is a disruptive enough change, it opens up a race between horizontal players (like Airtable and Recode) providing generic tools for building customized software and vertical SaaS players integrating deeper customization into their platforms.

While low-code players have moved to embrace these technologies first, I believe vertical SaaS products that successfully incorporate customization platforms (like Figma) will end up winning for a few reasons:

1. When you pay for a vertical SaaS, you're largely paying for domain modeling. This domain modeling is very useful even if you deeply customize the software on top of it. Salesforce is the #1 example of how this can be executed effectively: even companies, like Google, that typically prefer to build their own software pay for Salesforce. It's definitely not for the pretty UI; it's for access to a useful set of domain models that they can deeply customize.
2. Go-to-market for platforms is substantially more complicated than for vertical solutions. How would Airtable reach the buyer at my friend's non-profit? Even if they saw an ad for Airtable and tried the product, it would be hard to get them to see it as a real solution to their fundraising problem without clearer positioning.

Of course, there is one clear example of a company that built a massive GTM machine around a generic platform: Amazon. If low-code players want to seriously own the space of customized software, they'll need to embrace two things AWS does very well:

1. Build a serious marketplace that encourages independent vendors to build their domain models on your platform: No serious ISV can completely ignore Amazon's marketplace, while Airtable's marketplace still feels like an afterthought.
2. Provide deep, vertical-focused solutions. A few scattered case studies isn't sufficient: Amazon has tons of industry-specific messaging and solution architects.

Overall, I expect vertical players to still dominate SaaS, especially as a third category of companies crops up providing customization primitives as a service. (For example, Activepieces makes built-in integrations easier and Supertokens makes open source customizable auth.) There's still a massive opportunity for a company to provide powerful primitives around WebAssembly (maybe Fermyon) and code generation.

I'm looking forward to a day where there is no build vs. buy debate for most software.