The Great Decoupling: AI’s Impact on Software Economics and the "Demotion" of the Seat

Executive Summary

For the past twenty years, the "Software-as-a-Service" (SaaS) business model has operated on a simple, predictable equation: Value = Headcount $\times$ Price Per Seat. This model was a bet on human labor as the primary driver of business output, with software serving merely as a tool to facilitate that labor.

In 2026, we have reached an inflection point. Artificial Intelligence is fundamentally decoupling business value from human headcount. We are transitioning from a model of "Rent" (access to a tool) to a model of "Utility" (payment for work performed).

The Core Investor Dilemma:

Recent market volatility—termed by some as a "SaaSpocalypse"—has driven valuations for stalwarts like Salesforce from ~30x Free Cash Flow (FCF) to ~15x. This is not a revenue collapse; it is a terminal value reset. The market is no longer willing to underwrite 20–30 years of durable cash flows because AI introduces radical uncertainty into year 7 and beyond.

Key Investment Conclusions:

• ●SaaS is "Demoted," Not Dead: Mission-critical Systems of Record (Salesforce, ServiceNow) are not being "ripped out" for probabilistic AI code. However, they are losing their exclusivity on value capture.
• ●The "Agentic Layer" is the New Workspace: Value is shifting up the stack. Agents that orchestrate work across applications (e.g., pulling data from Slack, CRM, and email to execute a task) will capture the profit pool previously reserved for the applications themselves.
• ●Margin Structures are Resetting: The zero-marginal-cost era is over. Inference costs will compress gross margins, but this is offset by larger absolute gross profit dollars as software captures budget previously allocated to services.

Why This Debate Exists Now: The Terminal Value Reset

To understand the magnitude of the current shift, one must distinguish between Operational Failure and Valuation Reset.

Signal vs. Noise: The "SaaS is Dead" Fallacy

There is a prevalent but incorrect narrative that AI will immediately replace SaaS applications. On Today’ All-In Podcast, the besties had a vibrant discussion on this very topic. David Sacks notes, "You’re not going to want to replace [core systems] with code that was probabilistically generated yesterday.".

• ●Operational Reality: Revenue for major SaaS players remains stable or is growing.
• ●Valuation Reality: Stocks are crashing because investors are discounting future uncertainty.

Brad Gerstner (Altimeter Capital) argues that the compression in multiples (e.g., Salesforce dropping to ~15x FCF) reflects a loss of faith in long-term durability. "They’re going down because we’re discounting future uncertainty... We don’t know what happens seven years into the future anymore.".

The "Klarna Effect" as a Leading Indicator

The urgency of this debate is driven by tangible market signals. Klarna reported its AI assistant was performing the work of 700 full-time support agents, contributing to a 50% workforce reduction while revenue grew.

This is not a story of software efficiency; it is a story of labor replacement. When a company reduces its support staff by 50%, the revenue for its seat-based software vendors eventually declines. This lag creates a dangerous "air pocket" for investors: revenue looks stable today due to multi-year contracts, but the underlying user base is eroding.

The End of Infinite SaaS Leverage

The introduction of variable inference costs means that software companies are beginning to resemble lower-margin services businesses in their cost structures.

From "Helping Work" to "Completing Work"

David Friedberg highlights a subtle but massive shift in the core promise of software:

1. Old Model: Software helps a human do work (Productivity Tool).
2. New Model: Software completes the work for the human (Service Replacement).

This shift breaks seat-based pricing. If software completes the work, charging for a "seat" is illogical. Vendors must shift to outcome-based pricing (e.g., per resolved ticket, per closed audit), effectively absorbing the economics of a services firm.

Gross Profit Dollars > Margin Percentage

Investors must recalibrate. A lower margin percentage does not necessarily imply a worse business if the Total Addressable Market (TAM) expands.

• ●Scenario A (Traditional SaaS): Vendor sells a tool for $10k/year at 85% margin. Gross Profit = $8.5k.
• ●Scenario B (AI Operator): Vendor sells "digital labor" replacing a $50k employee for $20k/year at 60% margin. Gross Profit = $12k.

The Agentic Layer: The New Profit Pool

The most critical insight for 2026 is that Agents do not replace SaaS—they sit above it.

The Shift from App to Agent

Historically, the "Workspace" was the SaaS application (e.g., you lived inside Salesforce or Jira). In the AI era, the Agent becomes the workspace.

• ●The Threat: Users will interact with an AI agent (e.g., Claude, OpenClaw) that has permission to read/write across all underlying apps (Slack, Gmail, CRM).
• ●The Consequence: The SaaS application is demoted to "infrastructure"—a headless database that stores the record but loses the user interface and the primary value capture.

As Sacks argues, "SaaS becomes an old layer of the stack… and all the action moves somewhere else.". Investors should be wary of companies whose primary value is a "workflow interface," as this interface is being disintermediated by agents.

Where Value Accrues: The Stack Analysis

Not all software companies are equally exposed. We propose a three-layer framework to categorize vulnerability and opportunity.

The Kill Zone: Systems of Engagement & Thin Wrappers

• ●Definition: Interfaces for humans to input or view data (e.g., basic CRM wrappers, simple support desks).
• ●Analysis: These are "thin wrappers." If an AI agent can interact directly with the database via API, the human interface becomes redundant.
• ●Verdict: Avoid. These assets face existential risk. As noted in the podcast discussion, "If you’re just a thin application layer sitting on top of a CRUD database, you’re in trouble.".

The Battleground: Systems of Record

• ●Definition: The "Source of Truth" databases (Salesforce, SAP, Oracle).
• ●Analysis: These incumbents possess "Data Gravity" and deep trust. You cannot "hallucinate" a bank balance or an inventory record.
• ●Verdict: Hold, but Reprice. These companies will survive as the "system of record," but their multiple will permanently compress as they lose the "system of engagement" premium. They become the "plumbing" rather than the "building."

The Winners: Data Infrastructure & Systems of Action

• ●Definition: Platforms that store, transform, and feed data to agents (Snowflake, Databricks) and Vertical SaaS that owns the outcome.
• ●Analysis: AI depends entirely on clean, accessible data. Companies like Databricks are re-accelerating (60%+ growth) because they are the "fuel stations" for the AI engine.
• ●Verdict: Overweight. "Data platforms are beneficiaries of AI, not victims.".

Horizontal vs. Vertical Outcomes

The distinction between Horizontal and Vertical software is becoming the defining line between commoditization and durability.

Horizontal SaaS (The "Wrapper" Risk) General-purpose tools without data gravity (e.g., generic project management, note-taking apps) are highly vulnerable. Their features are easily subsumed by foundation models or OS-level agents. "Using Notion’s AI is nice... But when you build agents that pull from everything, it’s unbelievable.". The cross-app agent wins over the in-app copilot.

Vertical SaaS (The "Operator" Opportunity) Vertical companies are insulated because they own the Workflow. A generalist LLM does not know the specific regulatory nuances of a construction permit in a specific county. Vertical SaaS companies will transition from selling software to selling "completed work" (e.g., ServiceTitan), capturing the services revenue stream.

Competitive Advantage in an AI-Native World

The traditional concepts of "Moats" are being eroded and rebuilt.

The Real Moat: The Feedback Loop (RLHF)

Proprietary data is often overrated; general knowledge is free. The durable advantage is the Reinforcement Learning from Human Feedback (RLHF) loop.

• ●Mechanism: When a user corrects an AI agent, that signal is gold.
• ●Flywheel: This fine-tunes the model for that specific domain. Usage is the new moat.

Switching Costs: Contextual vs. Technical

AI lowers technical switching costs (agents can auto-refactor code). However, it raises contextual switching costs. If an AI agent has learned a company's unique voice and risk tolerance over two years, switching to a "fresh" agent is painful.

What the Market Is Getting Wrong

Misconception: "AI Replaces Software"

Reality: AI is replacing the human operator of the software. The software itself becomes more valuable, complex, and essential than ever. The "Seat" is not dead; it is being demoted to a billing metric for "Humans in the Loop," while the growth engine shifts to "Work in the Cloud".

Misconception: "Valuation Drop = Business Failure"

Reality: The collapse in SaaS multiples is a rational pricing of uncertainty. Investors are no longer willing to pay for 30 years of durability when the next 7 years are opaque. This is a "Terminal Value Reset," not a signal that Salesforce revenue is disappearing tomorrow.

Implications for Investors

For Public Market Investors (Long-Only)

• ●Stop Obsessing Over Seat Counts: Look for ARPU Expansion and RPO Growth. A flat seat count with 20% ARPU growth is a bullish signal of pricing power.
• ●Accept the New Multiples: Do not expect a return to 25–30x FCF for mature SaaS. The risk premium has permanently risen.

For Private Equity (PE)

• ●Due Diligence Revolution: You are buying "Model Weights" and "Data Rights." Audit for "AI Debt"—is the codebase written by humans or hallucinations?
• ●The "Service-as-Software" Play: Aggressively cut headcount in portfolio companies (support, QA) and replace it with AI efficiency tools to drive EBITDA.

For Venture Capital

• ●Seek the "Operator" Model: Invest in Vertical SaaS that owns the workflow.
• ●Avoid "Thin Apps": If the app is just a UI over a database, it will be eaten by an agent.

Conclusion: The Seat Is Demoted, Not Dead

The "Great Decoupling" is not a temporary cycle; it is a rewriting of the economic physics of the software industry. The "Infinite Leverage" of the zero-marginal-cost era is ending, replaced by a world of variable inference costs and high-value digital labor.

The Final Takeaway: It can be true that SaaS is not replaced—and also true that it never trades at 30x free cash flow again. The "Profit Pool" is shifting from the application layer to the Agentic Layer and the Data Infrastructure layer.

Investors must recognize that the "Seat" is not dead, but it has been demoted. The companies that cling to the 2015 SaaS playbook will fight a deflationary tide. Those that embrace the "Operator" model—selling results, managing inference margins, and owning the feedback loop—will build the giants of the 2030s.