Yes, AI Can

“This customer needs these new features” is perhaps the most ubiquitous phrase in vertical software. It haunts the builders of such software with the same vigor that it emboldens customers to lodge new requests. Between the customer and the engineers sits a cacophony of middlemen: product, support, success and sales teams. Their role is to append the customer’s request in unison before it lands on the desk of the engineers: “it’s urgent,” as if the backlog of urgent feature requests weren’t already several times greater than the team’s bandwidth. After all, the logic goes, if vertical software prides itself on serving one industry exceptionally well, these new features must be necessary.

Vertical software is denoted “vertical” because it serves only one industry. It seeks to become the one-stop-shop software for the industry by building a platform that provides all the software necessary to run a business in the industry. It continuously bolts on new features and products, known as vertical growth. Imagine a yoga studio only needing one software for all their class scheduling, accounting, marketing, and reporting. For most industries, this is a win: businesses need not adapt their modus operandi to the way a software works. By contrast, horizontal software, like a social network, aims to serve the whole world with one product.

Vertical growth can be a deathly trap for a software company. When customers sniff out that a vertical software exists to serve their industry, a tug-of-war begins. Not every yoga studio operates the same way. Soon enough, individual yoga studio owners will start raising specialized requests that uniquely suit their business, in ways discordant with the broader industry. At an inflection point, new incremental features do more harm than good. Rather than deepening the software’s vertical appeal, they start making the software untastefully clunky. Without intervention, the clunkiness compounds and customers are suddenly left with a software more Frankensteinian than valuable.

And that dance between specialization and clunkiness is the fundamental tenet of vertical software. If the industry that the company serves is a big enough market, the company can grow very large without overplaying its hand on specialization. In most cases, the first 80% of running a business in a particular industry is shared homogeneously. Focusing only on those shared needs and serving them exceptionally well is a common and successful tactic in vertical software, albeit a luxury reserved for larger industries. Yet even the largest vertical software companies are pressured, at least in part, to provide custom extensibility that specializes their software to the unique needs of their customers.

Historically, a company had only two ways to respond to incoming feature requests for bespoke extensions of their software. The first is to give customers the tools to build the features themselves. In larger industries, such as ecommerce, most customers will have the means to work with software development agencies who can build custom features on top of the vertical software platform using tools provided to them by the platform itself. This requires paying people to write and maintain custom code. Most industries lack the profit margins to support this kind of investment. Consequently, most vertical software platforms opt for the second option: to stack-rank the feature requests and begin implementing them, while purporting not to sacrifice either quality or usability of the broader platform.

The dawn of AI, specifically large language models (LLMs), presents a third, highly attractive option. Vertical software companies can build the features that serve the first 80% of functionality shared across all customers in an industry and use the resulting platform as scaffolding for an embedded LLM that allows customers to build their own features without needing to code. LLMs embedded in well-scaffolded vertical software can let customers type their bespoke feature requests in natural language and then autonomously build and maintain that functionality. Maintained autonomously by AI, the bespoke functionality can be gated to just that customer. As a result, the verticalized platform is extensible in an accessible fashion without breaking the bank or compromising usability.

The opportunity for vertical platforms to embed AI for bespoke extensibility is massive. In fact, it could spur into existence new kinds of hybrid platforms, which straddle the line between the horizontal and the vertical. Exceptionally tailored to industries yet equally malleable to the nuances of customer needs within those industries, these hybrid platforms will commingle the rigid feature-factory approach of the past with the feature-canvas approach of the future. Unlike software platforms that let users turn natural language prompts into any kind of website, the verticalized scaffolding will ensure customers don’t inadvertently build faulty code or in other ways jeopardize their experience. The rate of success will entirely depend on the quality of scaffolding and the reasoning capabilities of the underlying LLM.

Already today, LLMs are capable enough to generate sophisticated code. But the visionary genius will lie in building the right scaffolding around them. This is best understood when prompting smaller, less capable LLMs: ask them to multiply two large numbers and they will give you the wrong answer; ask them to write and run a piece of code that performs the same calculation, and you’ll get the correct answer every time. LLMs command symbolic languages impressively, but they need the right context. Embedding them inside the scaffolds of software platforms whose customers demand extensibility creates mighty symbiosis. This is how vertical software platforms of the future will evolve, without being bogged down by an endless stream of urgent feature requests or compromising the platform’s usability. Customers will rejoice.