Why SaaS Won’t Die: AI Turns Software into Intelligent Process Engines

From File Cabinets to Databases, Then to Working File Cabinets

Understanding AI's impact on SaaS starts with a 60‑year software evolution timeline. In 1960 IBM and American Airlines built the Saber system, moving paper reservation records into early databases. Subsequent milestones include electronic medical records and the 1987 ACT CRM.

"From 1960 to 2022, software history is simply turning file cabinets into databases." Yet databases alone did not make the world more efficient—security, account provisioning, and other overhead remain.

The AI era changes this: files become active. Tools like QuickBooks can now execute tasks autonomously instead of waiting for humans to retrieve data.

SaaS's Three Fates: Not All Software Will Die

Investor Alex proposes a three‑category framework that most market participants miss:

1. Seat = Output (High Risk) – If a seat is sold for producing work, AI can replace that work, making the seat obsolete. Example: Zendesk could see seat revenue vanish if AI‑driven customer service eliminates human agents.

"If Zendesk continues charging per seat without change, that revenue line will go to zero." Conversely, shifting to outcome‑based pricing could multiply revenue. 2. Seat ≠ Output (Safer) – Seats are a pricing mechanism, not a production unit. Workday charges per employee; the employee does not produce work within Workday, but AI can add value through automation (e.g., background checks) while the seat remains a fair‑feeling anchor. 3. Middle Ground – Companies like Adobe sit between the extremes; seats matter but are not as vulnerable as pure output‑based SaaS. Many investors fail to distinguish these categories, overlooking that AI must still run on underlying record‑system software.

Vibe Coding: Threat or Opportunity?

The "SaaS apocalypse" argument claims AI lets anyone code software, eliminating SaaS. Mike counters that building a Workday‑like system with Vibe Coding is terrifying. Alex cites David Ricardo’s comparative advantage: making a hamburger yourself is costlier than buying one, despite the ability to produce it. The real moat of software lies in decades‑long edge‑case knowledge, not the code itself. These deterministic rules, learned from experience, cannot be copied without the underlying expertise. For example, handling a maternity‑leave termination in Indiana is an edge case that Vibe Coding cannot solve without the embedded knowledge. Mike observes that Vibe Coding expands extensibility: internal teams can quickly build niche tools (e.g., a meeting‑room booking app) that still rely on Workday’s data, making Workday more sticky and valuable.

Pricing Fairness: Why Seat‑Based Billing Persists

The lock‑smith story illustrates that pricing is driven by perceived fairness, not cost. Seat‑based pricing feels fair because larger companies naturally pay more. AI‑driven consumption or outcome billing faces two pitfalls:

Consumption Billing – Casino Tokens : Customers cannot predict token consumption; vendors keep adding features that drain tokens, eroding trust.

Outcome Billing – Diminishing Returns : Initial cost savings are impressive, but each subsequent year the marginal saving shrinks, making the model unsustainable.

AI Product Bottleneck: Design, Not Technology

Users often ask AI chatbots for jokes, exposing a gap between capability and delivered value. The real challenge is design and experience: users need to know what AI is doing and trust its actions. Mike proposes a "50‑intern dilemma"—AI can do massive work, but it also generates endless questions that consume user attention. To earn trust, AI must ask for confirmation before acting while not becoming a nuisance, a new design problem. Atlassian’s approach blends document editing with a chat pane (75% document, 25% chat), akin to early mobile apps that evolved from simple web views to native interactions.

Conclusion: Enterprises Are Not Databases, They Are Process Networks

Mike ends with a key insight: enterprises are collections of coordinated processes, not static record systems. He splits processes into two types:

Input‑Limited (e.g., customer service) – AI adds cost‑saving efficiency.

Output‑Limited (e.g., R&D, marketing) – AI expands capacity and quality.

The future of SaaS lies in combining process orchestration with AI execution, turning software into indispensable intelligent workflow engines.