There’s a popular trope among VCs these days that “SaaS is dead.” At least one firm has declared “we’re not investing in software anymore.” What they’re typically referencing is traditional enterprise software, which enjoyed a two-decade heyday of high margins, predictable growth, and low churn. This category now faces the dual threat of AI coding (making it easy for customers to replicate many “legacy SaaS” products) and agentic competition (AI-native startups aiming to unseat the incumbents). The recent decline in public market multiples for SaaS companies is one measurable manifestation of a growing anxiety. Investors are signaling that they think/fear AI and AI labs will disintermediate the interfaces and disrupt the business models that made software such an appealing category to invest in for so long.
It’s irrefutable that the rise of Cursor, Claude Code, and Codex has changed software development forever. Over the last twelve months, what felt like a productivity hack for only the earliest stage of startups is now a full-on vibe coding revolution across companies of all sizes. App stores abound with AI generated submissions. Tech team job descriptions are converging faster than reporting lines with product managers, designers, and engineers all able to prototype, code, and synthesize customer feedback.
The bear case goes something like this (skip ahead if you’ve already heard it):
- Technology has been a moat for many years because it’s hard to hire the specialized talent to build it well, and once you’ve built it, it takes time for the next company to catch up. It may not be as defensible as, say, patent protection, but having a unique feature set created a 12-18 month mini-monopoly that in turn ensured a degree of pricing power.
- But in a world where anyone can write code, the new bottleneck is the collective imagination of your team, not the number of comp sci majors you can recruit. The value of having a headstart on your roadmap is likewise in doubt if a competitor's path to feature parity is weeks, not years, away.
- Worse still, enterprise software has traditionally priced based on number of seats or users. But agentic workflows are inherently about reducing the number of employees needed to fill those seats.
- All of this is happening at a time when the center of gravity is tilting toward an oligopoly of AI labs building everything apps – targeting consumers and engineers, and, increasingly, enterprise customers too. Originally, they were horizontal, but they’ve been ramping up vertical-specific teams, features, skills, evals, and model training procedures.
The gloom expressed by VCs and Wall Street obscures a more nuanced reality. The cone of uncertainty is wide. But the more likely scenario for the future of software is that our bar for “good enough” rises dramatically, our interaction patterns evolve, and when all is said and done, we end up using or buying more software (defined broadly) than ever before.
The history of technology bears this out – productivity gains reduce the marginal cost per unit of output but often (though not always) increase consumption (sometimes called a “rebound” effect or “Jevons Paradox” at the extreme). Companies could very well spend less on labor and more on software, because software will become the primary medium through which work gets done. They’ll continue to scoop up technical and semi-technical talent too. Why? Because as everyone’s development velocity increases, the equilibrium response isn’t to do less, it’s to outcompete by doing more.
What about the threat of moat erosion? It’s true that the rate of features shipped per developer per hour has increased enormously. But unless that rate ratchets up forever – far from guaranteed – the concept of a roadmap head start won't entirely disappear. Instead, we'll adjust our expectations of what can be achieved in a week or a month or a quarter. Companies that pick better problems to tackle, pair problems with smarter solutions, and execute with urgency will have an advantage – yes, even a durable one.
How will this change the software landscape? We should expect fewer point solutions and more “operating systems” or “super apps” that string together lots of use cases. The starting point a founder picks will matter a lot because it’s typically easier to go downstream than upstream in a workflow or series of processes. Platforms will combine agentic automation with human intervention and a mix of conversational and traditional UX. Third party point solutions will still exist in domains requiring extreme technical depth or complex system integration.
The concept of a "user" will change. In the future, software may have fewer users – but those users may be more authoritative. As AI consolidates work into fewer hands, the operators who remain in the loop are potentially a step or two closer to the buyer: managers, directors, executives. Less distance between the person who pays for the software versus uses it may actually smooth out sales cycles.
Monetization will evolve. As software becomes easier to replicate, pricing power will shift toward harder-to-copy capabilities – the parts of your product that are actually unique. That may mean more freemium pricing models or monetization tied to second-order benefits (e.g., marketplaces that connect buyers with sellers, access to networks of partners or peers). It will also mean a continued evolution from seat-based pricing to usage-based models. Will this result in a catastrophic decline in ACV? Not necessarily. Usage, or at least utility, will expand as software becomes more central to a company’s operations – as labor is usurped by automation.
Data, in particular, will become a strategic sticking point. We’re likely to see much greater focus on data rights in enterprise software deals, with contracts that explicitly define give/get terms or default to sharing de-identified insights across customers, up to and including the data needed to train agents on new workflows. In a world where interfaces and models commoditize, proprietary data remains a somewhat indestructible moat.
What about the incumbents? Consider this: CRMs have become a meme of throw-away software among vibe coders, yet OpenAI uses Salesforce as its CRM, as does Anthropic, as does Vercel. The fact is that many of today’s mature software businesses, especially systems of record, are in a position to shepherd their customers into the AI era rather than concede turf to AI-native challengers. The lesson for founders is to seize the AI moment, automating everything around the core systems as a foot-hold to replacing them and becoming as deeply embedded as they are. Also, the bigger the enterprise you build for, the more heightened the security and procurement posture, the lesser the risk of software commoditization.
In the end, will the AI mega labs own everything? It’s tempting to think so. But the surface area of complex industries like healthcare and fintech is so vast, their workflows so gnarly, true industry expertise so precious, and access to frontier models through APIs so ubiquitous – that many great companies will be built alongside them, formidable as they are. If horizontal AI is about nailing the first mile, building tech for pharma, hospitals, insurers, banks, and logistics networks is all about the last mile – how does your solution actually fit into the workflow and complete an end-to-end task? Which other systems must it touch and what do you do when they don’t have any APIs to speak of or are on prem and written in COBOL? How do you deal with PHI or PII, or for that matter state specific data privacy laws? What if the expertise you need to automate your work is nowhere to be found on the public internet? What if your solution triggers ONC requirements for clinical decision support? Or implicates SEC/FINRA rules on what constitutes investment advice?
So no – software isn’t dead (unless you want to get mired in semantics). But we are in a period of radical recalibration, and with news of Mythos, the fog of uncertainty is as thick as it will ever be. The companies that win will capitalize on new models of defensibility. They’ll re-organize their teams around rapid experimentation. They’ll set roadmaps for long-term differentiation knowing that timelines will collapse periodically. They’ll use AI to replace superfluous workflows and invent new ones, with the right human-UI-AI hand-offs. And in the corners of the economy with the greatest complexity and regulatory sensitivity, they’ll bridge the gap between messy operational reality and AI-enabled possibility.