The recent AI-inspired software meltdown is a gross overreaction. SaaS companies will do well. In fact, they may see their business booming as a result of AI.
Having worked in the IT and cloud fields for decades, I know what it takes to build, sell, and maintain complex enterprise software. And it’s clear that AI is not the threat that many investors fear.
Let’s take a look at the three main concerns the market is preoccupied with. All of them have drawbacks to sustain the performance of existing SaaS vendors.
Enterprise DIY
Software development has been transformed by AI. This is a near-perfect use case for generative AI, and applying established patterns to well-bounded use cases yields incredible productivity gains.
Some commentators predict that this coding revolution will lead companies to write their own software rather than buying it from SaaS vendors.
This makes me wonder if these commenters have actually worked in enterprise IT at all. Even if we accept that creating software is now much easier, bringing a complete software product to market requires more than code.
- Domain expertise (regulatory requirements, industry practices, supply chain expectations).
- Education (marketing, pre-sales support, prototyping).
- Support (upgrade help, customer use case enablement).
- Product management (selecting user personas to address, defining and prioritizing features, prioritizing customer groups to address).
- Financial negotiations (special deals on quantities, discounts for official approvals, etc.).
- Legal “inoculation” (compensation, public policy advocacy to shape process guidelines).
Geoffrey Moore, a prominent management consultant and organizational theorist, wrote a book called “Crossing the Chasm” about what it takes for mainstream companies to adopt new technologies. He never mentioned the cost of writing software code as a gate element.
Throughout my long career, I’ve seen countless DIY failures by corporate IT organizations who fail to understand the difference between an internal software project and an actual product. We can already see another wave of failures fueled by misguided enthusiasm for AI coding.
AI startup disruption
AI forecasters expect cheaper startups to replace established software giants. This overlooks the reality that established SaaS providers already have smaller, cheaper competitors and yet still somehow maintain a dominant position. The challenges for new entrants to the software field include:
- There are obvious problems inherent to the digital sector, such as network effects and lack of scale.
- Some are less obvious, such as the need for geographic vendor reach (“Is local native language consulting available in country XYZ?”).
- Enabling custom integrations for large enterprises. Large companies often require bespoke software configurations.
- Weird contract requirements that are difficult for small vendors to support. Again, large customers often ask for special treatment.
It is very difficult for small startups to replace established vendors. As Clayton Christensen wrote in “The Innovator’s Dilemma,” innovative startups typically begin by solving a use case that existing vendors cannot or do not want to address.
What is left unanswered in this scenario is why incumbents are not using AI within themselves to improve engineering efficiency and address pricing pressures from smaller new competitors.
go vertically
The idea is that AI model companies will expand their initial software products into vertical products, thereby obliterating incumbent vendors.
For example, OpenAI has made a lot of noise with its efforts in healthcare. Anthropic caused a drop in many software stocks with its plugins.
It’s understandable why these companies started this initiative. Many industry-specific software companies are highly profitable, and AI Lab wants to be in this business.
However, pursuing this while working on other initiatives can make your AI lab too dispersed. Startups often fail due to lack of focus, and this is especially true for AI modelers. They face a huge and unprecedented opportunity, and getting distracted by bright, shiny vertical SaaS products is a terrible idea.
Returning to the DIY section above, actual enterprise software is very complex and expensive to ship and maintain. Multiply that by all the different industries that exist, like healthcare, financial services, manufacturing, etc. Dealing with the unique requirements of each department requires the time and attention of managers, as well as a large number of employees. Model manufacturers are already growing at breakneck speed. Trying to grow enough talent to become a vertical software provider will be a Sisyphean task.
beanie baby
If I were to advise these committees, I would advocate for them to focus on winning the horizontal AI model layer, which is likely to become a small oligopoly.
And yes, expanding AI coding capabilities can reduce development costs and increase the world’s population of software creators. This dynamic can lead to the Jevons Paradox (cheaper software leads to more software) and can enrich model providers without forcing them into every software field.
Looking back, SaaSpocalypse can now be thought of in the same breath as Beanie Baby mania. It is a short-lived phenomenon that now seems incomprehensible.
Bernard Golden is the CEO of Navica, a Silicon Valley-based technology analysis, consulting, and investment firm.
