For most of my career, the economics of enterprise software were reasonably stable. Building was expensive. Maintaining was harder. Buying a subscription allowed an organisation to access capability it could not sensibly reproduce itself.
I am beginning to question how long those assumptions will hold.
The falling cost of software creation does not mean every organisation should build everything. SaaS products still provide scale, specialist knowledge, reliability and ongoing investment. But the original bargain deserves more scrutiny when AI can increasingly generate code, tests, interfaces and documentation from a clear specification.
My interest is practical. SaaS subscriptions are a material and growing cost. If we only use a narrow part of a broad platform, or if the vendor’s process forces avoidable work into our operating model, the equation may change. A tailored capability that once required a large team and a long project may become achievable with a much smaller engineering effort.
The question is not whether AI can produce a convincing demonstration. We already know it can. The question is whether the full lifecycle cost is lower after security, integration, data, performance, monitoring, support, regulation and change are included. Cheap code can still create expensive systems.
This is why I am interested in specification-led development rather than indiscriminate building. If the business can define the outcome, process, rules, constraints, and tests, AI can carry more of the build. Developers can then concentrate on the areas where professional engineering judgement matters most: architecture, model choice, performance, maintainability, security, and cost.
The strategic asset may shift from the code to the specification. A clear description of how the organisation intends to work can be used to generate, test, and regenerate the implementation as technology changes. That makes business architecture and process knowledge more valuable, not less.
I do not yet know how far this goes. Some subscriptions will remain far cheaper and safer than internal alternatives. Others may become difficult to justify. The useful discipline is to stop treating renewal as the default.
If software stops being the primary constraint, the harder constraint becomes organisational clarity. Can we describe what we need well enough to build it? That may be the question that reshapes both the SaaS market and the role of Digital.