Table of Contents
The narrative sweeping through Silicon Valley suggests that the era of Software as a Service (SaaS) is ending—that AI will simply "vibe code" replacements for every major enterprise tool, leaving incumbents in the dust. However, Anish Acharya, General Partner at Andreessen Horowitz (a16z), offers a contrarian and data-backed reality check. Far from a "SaaS massacre," the market is seeing incumbents strengthen their positions, raise prices, and leverage AI to solve problems that traditional software never could. In a wide-ranging discussion on the future of technology, Acharya dissects the real shifts in enterprise value, the rise of "weird" consumer apps, and why the "death of software" is vastly oversold.
Key Takeaways
- SaaS is not dead; it is evolving. Contrary to the "SaaS apocalypse" narrative, 75% of public SaaS companies have raised prices since the launch of ChatGPT, proving they still command significant leverage and value.
- AI lowers switching costs, ending the era of "hostage" software. Coding agents reduce the friction of system integration, forcing vendors to compete on product quality rather than locking customers in via complexity.
- Value aggregates at the application layer, not the model layer. As foundation models become commoditized substitutes (similar to cloud providers), the real value shifts to apps that orchestrate workflows and specialize in user experience.
- "Weird" wins in consumer AI. While big tech companies are constrained by safety committees, startups have a massive opportunity to build emotional, human-centric, and "weird" products that incumbents refuse to touch.
- The economics of AI demand a new framework. While AI apps may have lower gross margins due to inference costs, they benefit from lower Customer Acquisition Costs (CAC) and significantly higher willingness to pay from power users.
The Myth of the SaaS Apocalypse
The prevailing fear among investors is that traditional enterprise revenue is no longer sticky or durable. The theory posits that AI agents will simply rewrite code for payroll, ERP, and CRM systems, rendering giants like Salesforce or SAP obsolete. Acharya argues that this view fundamentally misunderstands how enterprises spend money and deploy technology.
Software currently accounts for only 8% to 12% of total enterprise spend. The idea of pointing an "innovation bazooka"—powerful AI models—at rebuilding a payroll system to save a fraction of that 12% is economically inefficient. Instead, the real opportunity lies in using AI to optimize the other 90% of operational spend that software has traditionally failed to touch.
You have this innovation bazooka with these models. Why would you point it at rebuilding payroll or ERP or CRM? The general story that we're going to vibe code everything is flat wrong and the whole market is oversold software.
Furthermore, the data contradicts the doom-and-gloom narrative. Since the release of ChatGPT, incumbents have not been forced to slash prices to survive. Instead, 75% of public SaaS companies have raised prices, with a mean increase of 8% to 12%. Price is a proxy for product-market fit; companies facing existential threats do not raise prices. Incumbents like ServiceNow are proving to be highly capable, integrating AI to reinforce their core advantages rather than being displaced by it.
From Hostages to Customers: The Drop in Switching Costs
While the incumbents are safe from being "vibe coded" out of existence, the dynamic between vendor and customer is shifting radically. Historically, enterprise software moats were built on high switching costs. Once a company implemented an SAP system, moving to Oracle was a multi-year, high-risk endeavor that could cost a CTO their job. This created a dynamic where vendors had "hostages, not customers."
The Role of Coding Agents
AI coding agents are dismantling this barrier. By automating the complex work of systems integration and data migration, agents dramatically lower the cost and risk of switching providers. This shift is a net positive for the ecosystem because it forces software vendors to maintain loyalty through product excellence rather than technical lock-in.
In this new environment, defensibility cannot rely solely on integration friction. It must return to fundamental network effects and systems of record that involve high-volume human workflows—areas where displacement remains incredibly difficult.
Why the Application Layer Wins Over Foundation Models
A critical debate in AI investing centers on where value will accrue: the foundation models (like OpenAI or Anthropic) or the applications built on top of them. Acharya compares the current trajectory of foundation models to the evolution of cloud providers. We are moving toward an oligopoly where models are roughly substitutes for one another, similar to how AWS, Google Cloud, and Azure compete.
Because models are becoming commoditized, the "aggregation layer"—the applications—becomes the strategic choke point. Applications provide value through:
- Orchestration: Tools like Cursor allow developers to seamlessly switch between models (e.g., using Gemini for the frontend and Claude for the backend) without managing multiple command-line interfaces.
- Specialization: Different workflows require different aesthetic or functional opinions. In creative tools, Midjourney offers a highly opinionated aesthetic, while other tools might prioritize neutrality for graphic design.
This creates a market structure where boring, broad applications don't necessarily win. Instead, specialized, workflow-rich applications that leverage multiple models will capture the customer relationship.
"Weird" Wins: The Frontier of Consumer AI
In the consumer space, Acharya introduces the thesis that "weird wins." Traditional technology has been quantitative and clinical. AI, by contrast, is unpredictable, emotional, and capable of simulating human connection. This creates a divergence between what big tech can build and what they will build.
What is the human experience? That often involves disagreement, persuasion, sexuality... Yet, if you're Google or Apple, you have a thousand committees that are explicitly designed to ensure there's never any persuasion, disagreement, or sexuality expressed in your products.
This corporate conservatism leaves a massive opening for startups to build products that explore the full spectrum of human emotion, including companionship, spirituality, and complex social interactions. Products like Replica or "contextual companions"—AI that plays Minecraft with a child to model pro-social behavior—address deep human needs for connection that large corporations deem too risky or "uncomfortable" to address.
Redefining Margins and Unit Economics
The transition to AI-native software requires investors to rethink their approach to margins. In the traditional SaaS world, high gross margins (80%+) were the gold standard. AI companies often face lower gross margins due to heavy inference costs (compute spend). However, Acharya argues that this structural drag is often offset by a more efficient go-to-market motion.
The New "Power User" Economics
In the pre-AI era, a power user of Spotify might pay the same $20 monthly fee as a casual user. Today, AI power users are willing to pay significantly more—often $200 to $300 a month—for tools that deliver tangible productivity gains. Furthermore, many AI apps grow via organic traffic (product-led growth), resulting in near-zero Customer Acquisition Costs (CAC) for the initial user base.
Investors should essentially view the initial lower margins as a form of "CAC spending." If a company offers a free tier that incurs compute costs but converts users into high-paying subscribers with high retention, the blended economics remain incredibly healthy.
The Future of Work: Agents and Ambition
Will AI agents replace human jobs? The reality is nuanced. Agents are exceptionally good at handling "low NPS work"—the repetitive, drudgery-filled tasks that humans generally dislike. However, they struggle with ambiguity. Most high-value work involves navigating uncertain contexts, making judgment calls, and handling exceptions.
The future of work, therefore, isn't about the replacement of the worker, but the expansion of the worker's ambition. When the cost of execution drops to near zero, the constraint on output shifts from capability to ambition. Humans will remain in the loop to handle ambiguity and direct the agents, effectively moving from "doing the work" to "orchestrating the outcome."
Conclusion
We are not witnessing the death of software, but rather its expansion into the vast majority of the economy that it previously couldn't touch. Whether it is through lowering the barrier to switching enterprise providers, creating new categories of emotional companionship, or redefining the economics of productivity, AI is an accelerant.
For founders and investors, the lesson is clear: don't bet on "vibe coding" existing solutions into oblivion. Bet on the "weird," the specialized, and the ambitious new categories that simply weren't possible before the current generation of models. As Acharya notes, the Net Promoter Score (NPS) of the human experience is on the way up, driven by technology that finally understands us.
