Software Is Dead — Long Live Software

The recent software sell-off is treated as evidence that SaaS is dying, but the piece argues that this reading is too broad. It says AI is still software, cloud-delivered AI is still SaaS, and the bigger shift is a new generation of software that changes which vendors gain value.

It also argues that cheaper software creation expands markets rather than shrinking them. Some hyper-narrow tools, LLM-subsumed products, and slow incumbents are exposed, but software businesses with proprietary data, regulatory lock-in, embedded workflows, and trust remain defensible. The text ends by saying that the next wave of value will come from vertical AI built on real-world coordination and data loops.

Reading notes

#

• A recente queda das ações de software is presented as a SaaS flash crash, but the text maintains that this does not prove the end of the model.
• The central argument is that AI remains software, and AI delivered in the cloud remains Software-as-a-Service.
• The text says that the paradigm shift brings new terminology and leads many people to confuse a new generation of products with the disappearance of the business model.
• The thesis is that AI makes software more accessible and easier to build, which expands the market.
• The text separates the most exposed companies into three groups: hyper-narrow tools, offerings that LLMs begin to absorb, and incumbents that are slow to adopt AI.
• It states that slow incumbents face the dilemma of cannibalizing the current business and tend to lose value to AI-native challengers.
• There is also the “front door” risk, in which agents capture the incremental value layer and push the system of record down the stack.
• Even so, the text says that software growth remains strong and that the cloud still has room to grow in enterprise workloads.
• The text uses the migration from on-premise to cloud as a precedent to show that technological shifts can kill some companies while also expanding the market.
• It states that software became more valuable when it became cheaper to build, sell, and maintain.
• The text highlights that what remains difficult is not writing code, but discovering the problem, designing the right workflow, distributing the product, and earning trust.
• The defense of software increasingly depends on domain expertise, distribution, customer relationships, and responsibility for outcomes.
• The text concludes that the software market grows with AI, but that value creation will shift toward startups and incumbents capable of building moats of data, workflow, and trust.