What started out as a cost saving exercise is turning into a battle against technical debt, says Jason Walsh
Blogs
Image: Christina Morillo via Pexels
Software is not content. It does not sit there waiting to be consumed. It runs, and when it runs it touches other things, and those things touch other things, and somewhere in that chain of consequences is your business.
By now, everyone will know that software shares plunged in February. Most of the time, Wall Street jitters are barely news: despite the tough guy rhetoric, the vast sums of money sloshing around in stock markets mean that even the faintest signal of disruption produces outsized reactions.
Nevertheless, the recent plunge in software company share prices – broad indices down roughly 25% from 12-month highs, with some names down considerably more – indicates a deeper concern: will artificial intelligence (AI) put an end to software, in particular to the now dominant subscription model of software-as-a-service (SaaS)?
The recent sell-off was triggered by release of AI-powered productivity tools by Anthropic. Investment bank Jefferies immediately dubbed it the SaaSpocalypse, but the underlying anxiety had been building for months.
To understand the fear, it helps to think about what SaaS actually is. SaaS solved a real problem in enterprise computing. It shifted the cost of software from capital expenditure to operating expenditure, moving it from paying huge lump sums and amortising over time to something more predictable and scalable. For businesses managing complex IT systems, this was a genuine improvement.
The prospect of generating custom applications cheaply and quickly brings us back to an earlier vision of the computer: the universal problem-solving machine, shaped around the problem rather than the vendor’s idea of how things should be done. But the question is not whether code can be written. AI code generation works well for discrete, bounded problems, but it becomes an obvious risk in the face of business-related interdependencies that mean doing something in one place can have consequences somewhere else that neither the user nor the model fully anticipates. The question is not whether AI can write software. It can. It is whether it should, and for what.
Morgan Browne, founder and chief executive of software company Enterpryze, said that much of today’s enterprise software was, while increasingly easy to use, doing complex things behind the scenes.
“It depends on how complex the software is. If it is writing a single function, fine. If you’re talking about something like an ERP system, how could it possibly understand that if it does something over here what that will do over there,” he told TechCentral.ie.
Business not tech
The upshot of this is that enterprise software needs to be auditable, and this reality is well understood by businesses.
“The reality is that the IT industry moves at 100 miles per hour and business moves at two miles per hour,” Browne said.
What businesses need, above all else, is certainty, and certainty requires accountability, which is as much a part of what software companies deliver. This was true even when SaaS overreached, notably in consumer software. The logic of the subscription, once liberated from the enterprise context that justified it, became a profit mechanism dressed as a convenience.
In the enterprise, though, SaaS remained a mechanism with a name and an address attached, unlike custom code developed without deep expertise that accumulates silently as technical debt, compounding with every update, as stranded assets when the person who built it moves on, and as a compliance liability when the auditors arrive. From compliance to bug squashing and from updates to customisation, trust is as much a part of the picture with software as is functionality.
In an increasingly complex regulatory environment – think about the Digital Operational Resilience Act (DORA) or Artificial Intelligence Act (EU AI Act) – chains of responsibility and clear transparency matter at least as much as the bits being shifted around. Software companies would be smart to point out precisely what it is they deliver, and it is not bits.
“I have people in support, in consulting, in selling and marketing,” Browne said. “What people are forgetting is you want someone to be responsible if something goes wrong. You want someone who will answer the phone.”
AI is already capable of impressive feats, but the alternative being posited to commercial software is not simply cheaper software, it is software without a responsible party, built by a process its users do not fully understand, maintained by no-one in particular, and answerable to the void.
In other words, precisely the problem free and open source (FOSS) software sought to fix decades ago. This requires not only human oversight but deep technical expertise, precisely what someone knocking out sketchily understood functions lacks.
Markets have priced in disruption. They have not yet priced in the question of who picks up the phone. The most important thing that SaaS companies can do now is point out that subscriptions came with contact details.