Over the past months we have seen a brutal repricing of software-related equities. Companies like Shopify, Salesforce and Adobe have all been swept into a broad sell-off, often treated as if they belong to the same fragile category: "software threatened by AI". The narrative is seductive, simple and, in my view, dangerously incomplete.
A significant trigger has been the rapid progress from Anthropic, most notably the release of new versions of Claude Opus. Demonstrations show systems trained to excel in narrow but demanding domains like law, medicine and scientific reasoning. On top of that, Anthropic is pushing hard on multi-agent collaboration and has released agent orchestration in beta. The implied message to the market seems to be: "Why buy software when you can just build it yourself with AI?"
This idea is powerful, but it is also deeply misleading.
The productivity gains are real
Yes, the productivity gains are real. Over the last three months alone, the practical usefulness of Anthropic's models has increased at an exponential rate. Tasks that previously required teams can now be prototyped by individuals. Code can be generated, refactored and tested at speeds that would have seemed absurd a year ago. From a pure capability perspective, this is not incremental change, it is a phase shift.
But capability is not the same as responsibility.
SaaS is not just features. It is accountability
Modern SaaS platforms are not just bundles of features. They are operational backbones. They encode years of edge cases, compliance work, security practices, uptime guarantees and battle-tested implementations. When a business runs on Shopify, Salesforce or Adobe, it is not only buying functionality, it is buying accountability. There is a vendor that owns the system behaviour, patches vulnerabilities, documents changes and can be held contractually responsible when things go wrong.
AI-generated code does not come with that guarantee.
If an organisation decides to replace stable SaaS components with bespoke, AI-written systems, the responsibility does not disappear. It moves entirely in-house. There is no external party to hold accountable. No vendor roadmap. No shared liability. Every bug, security flaw or architectural mistake becomes the organisation's own problem. This is not a theoretical concern. It is an operational reality that many decision-makers underestimate.
The market is overreacting
This is where I believe the market is overreacting.
Index-following capital and short-term momentum traders tend to flatten nuance. Software becomes "software", and software becomes "at risk". But the idea that all SaaS value evaporates because AI can generate code ignores how real system landscapes function. Enterprises do not optimise for novelty. They optimise for stability, predictability and risk management. That backbone matters more, not less, as the surrounding pace of change accelerates.
Disposable code has a place. But not everywhere
There is, however, a crucial distinction to be made.
AI-generated code absolutely has a place. But that place is best framed as disposable code. Code that accelerates exploration, experimentation, internal tooling and non-critical workflows. Code that is allowed to be replaced, rewritten or thrown away when assumptions change. When paired with strong processes, guardrails and clear ownership, this can unlock enormous value without undermining the core architecture.
What is dangerous is confusing disposable code with foundational infrastructure.
More power demands more responsibility
The irony is that the more powerful AI systems become, the more important clear architectural responsibility becomes. Multi-agent systems and orchestration layers increase leverage, but they also increase complexity. Without well-defined system boundaries and ownership models, organisations risk building landscapes that are fast to create and impossible to govern.
I would argue that very few people truly understand the full implications of what is happening right now. The technology is moving faster than the mental models used to evaluate it. As a result, we see "dumb money" reacting to headlines rather than fundamentals, pricing fear instead of structure.
SaaS plus AI, with clear lines
Looking ahead to 2026, one thing feels certain: productivity will continue to explode. The winners will not be those who rip out their SaaS backbone in a rush to chase AI novelty. The winners will be those who combine proven platforms with AI-driven acceleration in a deliberate, responsible way.
The future is not SaaS versus AI. It is SaaS plus AI, with clear lines of accountability. That distinction is easy to miss in a sell-off, but it is where long-term value will be created.
And yes, it is going to be genuinely exciting to see where we land.