The recent AI-inspired software collapse is a huge overreaction. SaaS companies will do just fine. In fact, they could see their business flourish thanks to AI.
Having worked for decades in the IT and cloud industries, I know what it takes to build, sell, and maintain complex enterprise software. And it’s clear to me that AI is not the threat many investors fear.
Let’s look at the three main concerns that are obsessing the market. All have flaws that will remain established SaaS sellers are doing well.
DIY business
Software development has been transformed by AI. This is a near-perfect use case for generative AI: applying established models to a narrowly defined use case, resulting in incredible productivity improvements.
This coding revolution has led some commentators to predict that companies will write their own software rather than purchasing it from SaaS providers.
This makes me wonder if these commenters have ever spent time working in enterprise IT. Even granting that building software is now much easier, bringing a complete software product to market requires much more than code:
- Domain expertise (regulatory requirements, industry practices, supply chain expectations).
- Training (marketing, pre-sales support, creation of prototypes).
- Support (upgrade help, customer use case enablement).
- Product management (selection of user profiles to be addressed, definition and prioritization of features, prioritization of customer groups to be addressed).
- Financial negotiation (special offers for volume, discounts for public approval, etc.).
- Legal “immunization” (compensation, public policy advocacy to shape process guidelines).
Geoffrey Moore, a renowned management consultant and organizational theorist, wrote a book called “Crossing the Chasm” about what large companies need to adopt new technologies. He never mentioned the cost of writing software code as a determining factor.
During my long career, I have witnessed countless DIY failures from corporate IT departments who failed to understand the difference between an internal software project and an actual product. I’m already seeing another wave of failures, fueled by misplaced enthusiasm for AI coding.
AI startup disruption
AI prognosticators see cheaper startups supplanting large incumbent software companies. This ignores the fact that established SaaS providers already have smaller, cheaper competitors, but still remain dominant. Challenges for new entrants in the software sector include:
- The obvious problems inherent in any digital sector, such as the lack of network effects and scale.
- Less obvious questions, like the need for geographic reach of the provider (“Do you have local advice in your native language in country XYZ?”).
- Possibility of custom integration for large companies. Large companies often require custom software configurations.
- Strange contractual requirements, difficult to meet for a small supplier. Again, large customers often want special treatment.
It is incredibly difficult for a small startup to displace an incumbent supplier. As Clayton Christensen observed in “The Innovators Dilemma,” innovative startups typically start by solving use cases that incumbent vendors can’t or won’t serve.
Not addressed in this scenario is why incumbents would not use AI themselves to improve engineering efficiencies and respond to any pricing pressure from new, smaller competitors.
Go vertical
This is the idea that AI model companies will expand their nascent software products into vertical offerings, thereby killing off incumbent vendors.
OpenAI caused a sensation with a health initiative, for example. Anthropic caused many software stocks to fall with its plugins.
We understand why these companies have launched these initiatives. Many industry-specific software companies are very profitable, so AI labs would like to have a piece of this business.
Pursuing this approach, while working on other initiatives, could, however, disperse AI laboratories too widely. Startups often fail due to a lack of focus, and this is particularly relevant for AI model creators. They’re facing huge, unprecedented opportunities, and getting distracted by shiny, shiny vertical SaaS offerings is a really bad idea.
Going back to the DIY section above, shipping and maintaining true enterprise software is extremely complex and expensive. Now multiply that number by all the different verticals that exist, such as healthcare, financial services, and manufacturing. Meeting the unique requirements of each sector would require a considerable number of employees, as well as management time and attention. Model makers are already experiencing dizzying growth; trying to add enough people to become vertical software providers would be a Sisyphean task.
Baby Hat
If I were advising these boards, I would argue for focus: winning the horizontal layer of the AI model in what is likely to become a small oligopoly.
And yes, develop AI coding capabilities that reduce the cost of development and increase the global population of software creators. This dynamic could trigger Jevons’ paradox — cheaper software that generates a lot more — enriching model providers without forcing them to jump into every software sector.
The SaaSpocalypse will, in retrospect, come to be seen as the Baby Hat mania: an ephemeral phenomenon that now seems inexplicable to understand.
Bernard Golden is CEO of Navica, a technology analysis, consulting and investment firm based in Silicon Valley.