AI Apocalypse: What, Me Worry?

Plant rave at the greenhouse

I worked for a large software company and implemented one of the company's products for internal use. When I started with the company just a few years ago, executives would talk about building something of value for the long term and not looking only at the short term. Recently, however, there was a panicked flurry of town halls promoting education in and use of agentic AI, or LLM models trained to follow directions rather than conversing. And then came the Reduction in Force (RIFs): 30,000 people laid off by email. I was not surprised to be included. After all, I implemented software for a software company which has come to doubt the value of software.

Software: from product to SaaS to ????

At one time software was sold as a product. Business customers would buy the product to use and then would buy it again when they needed updates. They might have bought a support package. Or they would build their own software using open source solutions. After this period, there was software as a service (SaaS). During this time, companies would charge rent for the software and deliver updates quarterly or more frequently. The drawbacks of these business models was that customers would pay for generic software and then customize it to their needs. Much of the improved features of SaaS were driven by generic customer need and not to the specific needs of each customer. So customers were paying for the development of features that helped SaaS companies sell their solutions to new customers, but having to bridge the gap for their own processes at their own expense. For example, a donor-advised fund organization might buy an off-the-shelf ERP system and bridge the gap by creating a report to rebalance investment pools, which their ERP system doesn't do. Total cost of ownership became: pay rent and also pay internal developers to make customizations.

The existential fear of SaaS companies is that with agentic AI, and with very little specialized knowledge, customers can build software to meet their own needs. Or, more likely, upstarts can offer this customization cheaply and chip away at big tech's market share. Because of this impending threat, SaaS companies need to cut costs by reducing headcount. Existing SaaS customers will continue to pay the same rent on something that SaaS has shifted resources away from: paying the same rates for less value. Vendors are cutting customer support, product development, and internal validation. As a trade-off, customers will receive AI features that promise to deliver the specialized functionality and product support that they need. While SaaS vendors are not lowering prices, they are rapidly changing how they build their solutions and so customers will need to evaluate if they are receiving the same value. One of the reasons that open source software was never really popular is that customers using it had to have their own software team customize everything heavily. When a company pays Microsoft rent for Excel, for example, they are paying for stability, backward compatibility, and future proofing. It will be interesting to see if Excel can retain its value as Microsoft changes how it is built and maintained.

We live in unprecedented times and nobody knows the future.

Mathematician Hannah Fry recently compared the advent of AI to the Y2K crisis, which was averted by serious attention and hard work. As SaaS undergoes its identity crisis, this will result in downstream impacts for every industry which depends on SaaS. A few years back, the slogan was that sofware was eating the world. Every field, no matter how blue collar or physically grounded, has integrated software into its operations. There are many questions about the hidden cost of AI: AI is very expensive in terms of power demand, water demand, and demand for training data; and it's being heavily subsidized in terms of its actual cost. In addition, this tool which enables rapid development is generating a lot of technical debt. There's a hope that technical advances will pay off in cheaper costs for AI but that's not guaranteed.

AI is disruptive not because it delivers more value but because it cuts costs (hat tip to Rory Sutherland ). As a consequence, corporations with short-term vision will need to adopt it heavily to remain competitive even if it means having to pay more over the long run. For those less constrained by short-term goals, who are interested in discovering and building value, there will be opportunities amid the peril.

What now?

Even though corporations are panicking, I follow the advice of Ignatius of Loyola who said "In time of desolation never to make a change; but to be firm and constant in the resolutions and determination in which one was the day preceding such desolation, or in the determination in which he was in the preceding consolation." In times of crises, people also need to especially vigilant against opportunists promising easy answers about things that nobody knows. For me, the firm resolution has always been that technology must serve human needs and the needs of the business. With such a resolution, AI often looks very much like a solution in search of a problem. Customers navigating this uncertainty must evaluate how closely solutions meet their needs now. This resolution also means that customers must be clear eyed about continuing to pay rent to companies that no longer believe in what they originally built and which may have decreased stability. And finally, customers need to be wary of promises which sound too good to be true, especially when those promises depend on things which have yet to happen.

I wrote this for myself, and the thoughts are my own, based on my experiences with technology. At no point did I use generative AI for brainstorming, "research," composition, editing, or in any other way.