Counter-Analysis

The Bear Case for Enterprise SaaS

Response to: Stefan Waldhauser, "Why AI Is Not Eating Enterprise Software"
Date: 2026-02-06
Method: First-principles analysis with adversarial stress-testing

What Waldhauser Gets Right

Before dismantling the weak points, credit where it's due.

1. Regulated transactional Systems of Record are genuinely defensible. SAP's general ledger at BMW, Epic's EHR at Mayo Clinic, Oracle's financial close process at JPMorgan — these store data that cannot be reconstructed, carries legal/regulatory standing, and has no "ambient signal" equivalent. AI agents will orchestrate atop these systems, not replace them.

2. Enterprise switching costs are real and measured in years. The Salesforce-Siebel transition took 7 years with a clearly superior business model. Data gravity — 10+ years of structured transactions, custom fields, and workflow configurations — creates genuine migration friction.

3. Current AI adoption is embryonic. Only 8.6% of enterprises have AI agents in production. 42% of executives report AI initiatives "tearing companies apart" organizationally. The 95% failure rate for AI workflow tools reaching production is not a sign of imminent disruption.

Waldhauser's thesis is directionally correct for regulated, transactional SoRs in large enterprises. His error is in generalizing this defensibility to all enterprise software — and in choosing Monday.com as his case study.

The Four Counter-Theses That Survived Scrutiny

I started with six counter-theses. Four survived scrutiny. Two I killed myself — one for being unfalsifiable, the other for circular reasoning. Each surviving thesis was pressure-tested against the strongest possible defense of Waldhauser's position. Here's what held up.

Counter-Thesis 1
Monday.com Is the Wrong Case Study — Workflow SoRs Are Fundamentally Different From Transactional SoRs

Not all Systems of Record are equal. The defensibility of a SoR depends on two properties of its data: irreplaceability (can this data be reconstructed from other sources?) and canonical authority (does this data have regulatory/legal standing?).

SoR Type Data Character Reconstructable? Regulatory? AI Vulnerability
Transactional
SAP, Oracle, Epic
Ledger entries, BOMs, patient records No Yes Low
Workflow
Monday, Asana, Jira
Task status, project state, assignments Partially No High
Communication
Slack, Teams
Conversations, decisions No (unstructured) Rarely Medium

Monday.com's core data — task assignments, status updates, project timelines, workload distribution — overlaps significantly with signals already present in communication tools, calendars, code repositories, and document activity.

"If I delete my Monday.com board, what algorithm rebuilds the task dependency graph from Slack messages?" Fair — and the thesis also lacks a clear falsifiability condition: what percentage of reconstruction counts as "enough"?

The refined claim isn't that AI perfectly reconstructs Monday.com's data. It's that the value Monday.com provides — coordination visibility and workflow orchestration — can be delivered by AI agents that observe and synthesize the same underlying work activities that humans currently report manually into Monday.com. The data doesn't need to be reconstructed; the need for a separate coordination tool diminishes.

AI agents observe ambient work signals → provide coordination visibility without a separate tool → workflow SoRs lose their exclusive function → pricing power erodes
Counter-Thesis 2
Per-Seat to Consumption Pricing Is Structurally Deflationary

Per-seat pricing works because each human user represents roughly constant value to the vendor. AI breaks this arithmetic in two ways:

Seat elimination: An AI agent replacing 3 humans doing project coordination eliminates 3 seats. Even if the organization pays for "AI consumption," the total spend is lower because the economic activity that justified high per-seat pricing is being automated.

Consumption price transparency: Once pricing is usage-based, it becomes directly comparable across vendors. This transparency accelerates commoditization pressure.

AWS and Snowflake maintain strong pricing power in consumption models. Consumption economics work differently when you control real infrastructure scarcity (compute, storage) versus "fake scarcity" via seat licenses.

Adobe generated $125M in standalone AI revenue in Q1 2025. The SaaS market grew from $266B to $315B — companies are monetizing AI, not cannibalizing margins.

This objection is valid for infrastructure-layer consumption (AWS, Snowflake) where real resource scarcity exists. But SaaS application-layer consumption is different — the marginal cost of an additional API call to Monday.com approaches zero.

The structural deflation isn't about consumption racing to zero. It's about the economic activity that justified per-seat pricing (human coordination work) shrinking while AI consumption grows but at lower total revenue per customer.

AI automates work that justified seats → seats decline → consumption pricing generates less total revenue per customer → per-customer economics deflate even if usage volume grows
Counter-Thesis 3
AI Enables Christensen Disruption From Below

AI doesn't need to replace SAP at BMW. It needs to make it economically viable to build domain-specific enterprise tools for the 10,000 mid-market companies that can't afford SAP today.

AI collapses the two biggest costs of enterprise software: development cost (AI-assisted coding reduces build time 3-10x) and configuration cost (AI agents eliminate consultant armies).

"80% of SoR value is surviving 10 years of edge cases, compliance requirements, and integration. An AI-powered mid-market tool will die the first time it hits SOC2 audit requirements."

And Salesforce took 7 years to displace Siebel even with a clearly superior business model.

The timeline objection is valid — this is the slowest-moving threat. But here's the thing: even granting a 10-15 year disruption window, conceding the timeline validates the directional thesis. If we agree disruption happens but debate the timeline, the investment implications are about when multiples compress, not whether.

The compliance objection is real for regulated industries but overstated for the broad mid-market. A 50-person marketing agency doesn't need SOC2-compliant project management.

Historical validation: Salesforce vs. Siebel, Shopify vs. enterprise e-commerce, Stripe vs. enterprise payments. In every case: incumbent kept the top of market, disruptor captured the growth market below.

AI lowers cost of domain-specific tools → new entrants serve underserved mid-market → incumbents retain enterprise but lose growth market → TAM compression at current multiples
Counter-Thesis 4
The Intelligence Layer Captures the User; the SoR Becomes Necessary But Commoditized Infrastructure

In technology transitions, the layer closest to the user's decision-making context captures the majority of value. AI agents are becoming that layer — synthesizing across multiple SoRs, presenting decisions in natural language, and acting on behalf of the user.

This thesis took the heaviest fire. SoRs aren't data transit layers — they're state authorities. Google didn't make websites "dumb pipes." And there's survivorship bias in the analogies: Google, iOS, and cloud are all cases where new infrastructure was created, not where existing SoRs were displaced. Oracle survived the cloud transition. AWS didn't kill Oracle.

Meanwhile, AI revenue (Adobe's $125M) is accruing within existing SaaS stacks, not to a separate intelligence layer. The intelligence layer isn't displacing the SoR — it's being absorbed by it.

The original thesis — "SoRs become dumb pipes" — is too strong. The refined claim: SoRs remain necessary but lose their monopoly on the user relationship. The intelligence layer sits atop and aggregates across multiple SoRs. Pricing power weakens because they're no longer the primary interface.

This is closer to telecom carriers than websites. Carriers didn't go to zero — they became lower-margin infrastructure. AT&T still exists. But value capture shifted to Apple and Google.

Acknowledged limitation: The timing is genuinely uncertain. Current AI agents are not yet reliable enough to serve as the primary enterprise decision surface. This requires AI reliability to reach 99.9%+ for critical workflows, which may be 3-7 years away.

AI agents aggregate across SoRs → become primary user interface → SoR vendors compete on API quality, not UX → pricing power shifts to intelligence layer → SoRs retain revenue but at lower margins

The Monday.com Problem

Waldhauser's choice of Monday.com as his case study is the weakest link in his argument.

01

Workflow SoR, Not Transactional

Its data overlaps with signals in communication tools, calendars, and code repos. Unlike SAP's ledger, Monday.com's data doesn't have the irreplaceability that makes transactional SoRs defensible.

02

Growth Proves Momentum, Not Defensibility

32% YoY growth is impressive. But BlackBerry grew 40% YoY in 2007. Siebel was still growing when Salesforce was a $176M company. Growth tells you about now, not about substrate shifts.

03

Already Disrupted by Simplification Once

Monday.com replaced MS Project and Primavera with simpler, flexible "boards." The tool that won by simplification is especially vulnerable to the next round of simplification — which AI enables.

04

Pure Seat-Based Pricing

Unlike Snowflake (real infrastructure costs) or SAP (regulatory lock-in), Monday.com's per-seat pricing has no structural defense against AI-driven seat elimination.

The stronger case study Waldhauser should have used: SAP S/4HANA, Oracle Cloud ERP, or Veeva Systems (life sciences CRM with regulatory data). These have irreplaceable transactional data, regulatory standing, and genuine data gravity.

The Strongest Version of the Bear Case

The bear case for enterprise SaaS isn't "AI replaces Systems of Record." It's more nuanced and more dangerous:

AI makes Systems of Record necessary-but-commoditized infrastructure.
1

Transactional SoRs survive but at lower multiples

SAP, Oracle, and Workday retain data authority but lose their monopoly on the user relationship. They become the "AT&T" of enterprise software — essential infrastructure, generating revenue, no longer commanding premium growth multiples.

2

Workflow SoRs face existential competition

Monday.com, Asana, and similar coordination tools face the most direct threat. Their core value proposition — giving humans visibility into work — is precisely what AI agents do natively. The data is reconstructable; the coordination function is replaceable.

3

Per-customer revenue deflates structurally

Whether through seat elimination, consumption pricing pressure, or Christensen disruption from below, per-customer economics compress. Total market may grow, but revenue per customer declines.

4

The timeline is measured in presidential terms, not quarters

The most important nuance: this disruption is real but not imminent. 8.6% AI production adoption, 95% failure rate for AI workflow tools, and 7-year displacement precedents all suggest a 5-10 year transition window.

What This Means for Investors