McKay Wrigley posted a thread in late March that generated more genuine finance-adjacent panic than anything I've seen in years. The premise was simple: he cancelled seven B2B SaaS subscriptions — Notion, Figma, Zapier, and four others totaling roughly $7,500 annually — after spending a weekend building AI-native replacements using Claude.
The reactions split predictably. Developers celebrated. SaaS founders panicked. VCs posted nuanced takes that mostly amounted to "this isn't that big a deal." The discourse missed the actual story, which is less about one engineer's weekend project and more about what happens when this behavior pattern scales across the 70 million knowledge workers who currently represent the SaaS industry's revenue base.
What Actually Happened
Before getting to the implications, it's worth being precise about what Wrigley built and what he didn't.
He did not build production-grade alternatives to enterprise Figma, Notion, or Zapier. He built personal workflow tools — custom dashboards, document management scripts, and automation pipelines — that replaced his specific use cases for those products. The distinction matters, but maybe not in the way SaaS defenders think.
The argument from SaaS bulls is that enterprise software has features, security, compliance, collaboration, and institutional knowledge baked in that a weekend Claude project can't replicate. That's true. It's also true for the bottom 40% of every SaaS company's customer base — the individual contributors, freelancers, small teams, and solo operators who are paying for enterprise-grade products but only using 20% of the features.
Those are the customers leaving first. And they have always been the hardest customers to keep.
The Churn Rate Problem
SaaS valuations are built on a fundamental assumption: that customers who have adopted a software product will continue paying for it. Annual net revenue retention (NRR) is the metric that drives everything — the DCF assumptions, the ARR multiples, the growth projections that justify paying 10x revenue for a SaaS company.
Best-in-class SaaS businesses have NRR above 120%, meaning they grow revenue from existing customers through expansion even as some customers churn. The average SaaS company has NRR in the 100-115% range. These numbers have been remarkably stable for the better part of a decade, and SaaS valuations have been priced accordingly.
The Wrigley moment represents a new source of churn that existing models don't capture: replacement churn.
Traditional SaaS churn happens when a customer stops needing a product's function entirely, or switches to a competitor offering the same product cheaper or better. Replacement churn is different — it happens when a customer builds a custom alternative that serves their specific needs better than the general-purpose product, at near-zero marginal cost.
Before AI, replacement churn was limited to engineers who could build custom software. That was maybe 5-10% of any SaaS product's user base. The vibe coding revolution — building software through natural language prompts to AI systems — extends replacement capability to the other 90%.
This is not a hypothetical future trend. It is happening now, and the NRR numbers that SaaS companies report with a quarter lag will not capture it until it's already a significant trend.
Which Categories Die First
Not all SaaS is equally vulnerable. The key variable is the ratio of generic workflow to proprietary data or network value.
Highly vulnerable (next 18 months):
-
Workflow automation (Zapier, Make, n8n): These products are pure orchestration — connecting inputs to outputs with rules and logic. This is precisely what LLMs and AI agents do natively. The replacement friction is near-zero for technical users and falling rapidly for non-technical ones.
-
Document and knowledge management (Notion, Confluence, Coda): The core value proposition — organizing and retrieving information — is now a prompt away with AI systems that can index, search, and surface information from any format. The collaboration features remain differentiated, but the fundamental product is replicable.
-
No-code/low-code tools (Webflow visual builder, Bubble, Airtable scripting): The premise of these products was that non-technical users could build software without code. AI chat interfaces now enable the same result with less friction for more use cases.
Moderately vulnerable (2-4 years):
-
Business intelligence and analytics (Looker, Tableau, Metabase): Data visualization and reporting workflows are increasingly replicable with AI agents that can query databases and generate visualizations on demand. The enterprise data governance layer retains value; the presentation and query interface does not.
-
Project management (Jira, Asana, Linear): Coordination infrastructure has deep integration value and switching costs, but the workflow layer — templates, automations, reporting — is increasingly commoditized.
Lower vulnerability (structural moat):
-
Vertical software with deep workflow integration (Veeva in pharma, Toast in restaurants, Procore in construction): These products have embedded themselves in industry-specific processes in ways that AI cannot replicate without the industry context.
-
Products with network effects (Slack, GitHub, Figma's collaboration layer): The value is in the shared workspace, not the underlying software. You can't replace Slack's organizational graph with a Claude project.
-
Data infrastructure (Snowflake, dbt, Databricks): The compute and storage layer of the data stack is not replaceable with prompts.
The Revenue Recognition Complication
Here's where this gets interesting for accountants and auditors.
SaaS revenue recognition under ASC 606 is relatively clean when customers pay a fixed subscription fee for access to a defined set of features. Revenue is recognized ratably over the subscription period. Contract modifications — upgrades, downgrades, cancellations — are handled according to specific guidance in the standard.
The emerging model — AI usage-based pricing with no fixed subscription — breaks almost every assumption embedded in SaaS revenue recognition.
If a company like Zapier shifts from a $50/month subscription to a usage-based model where customers pay per workflow execution, the revenue profile changes dramatically:
- Revenue is no longer ratably recognizable over a period
- Variable consideration estimates become necessary under ASC 606-10-32-5
- Contract term and termination rights change the performance obligation analysis
- Commission and contract cost amortization calculations change fundamentally
Every SaaS company accelerating toward usage-based pricing — partly to compete with AI, partly because their own AI features are priced per-query — is creating revenue recognition complexity that their existing accounting infrastructure is not built to handle.
The downstream effect on financial reporting: quarterly revenue becomes significantly harder to predict, analyst estimates will be systematically wrong during the transition period, and investors will demand new metrics to replace ARR and MRR that were built for subscription economics.
The Valuation Reckoning
The public SaaS market peaked at roughly $2.5 trillion in market cap in late 2021. The correction brought it down dramatically, and multiples have partially recovered. But those multiples still embed assumptions about churn rates and NRR that are built on a pre-AI competitive environment.
The SaaS Price-to-Revenue multiple for high-growth public SaaS companies currently sits around 8-12x. That multiple is justified by the assumption that the SaaS customer base is sticky — that once a company adopts a SaaS product and integrates it into workflow, they stay for years and often expand.
If replacement churn introduces structural NRR degradation of even 10-15 percentage points across the category — taking best-in-class NRR from 120% to 105% and average NRR from 110% to 95% — the DCF impact is severe. SaaS valuations are highly sensitive to long-run growth assumptions, and NRR is the primary driver of those assumptions.
A company with 95% NRR is not growing from its existing customer base; it's treading water and needs new customer acquisition to grow at all. That's a fundamentally different business than a 115% NRR SaaS company, and it should trade at a dramatically different multiple.
The market has not priced this in. Partly because the NRR data doesn't show it yet — churn is a lagging indicator, and the AI replacement wave is still in early innings. Partly because SaaS investors are structurally motivated to believe the category remains durable. And partly because the AI companies eating the SaaS market are not yet public, so there's no market signal forcing a revaluation.
What Actually Survives
The SaaS companies that will thrive through this disruption share one characteristic: they are building the AI layer, not just surviving it.
The clearest example is Salesforce, which has aggressively positioned its Einstein AI features as the primary value driver for enterprise customers. Whether the product actually delivers on the promise is debatable; what isn't debatable is that Salesforce understands the game. CRM data is the asset. The interface to that data — whether it's a traditional GUI or an AI agent — is just a delivery mechanism.
The SaaS companies that survive are the ones that own proprietary data that AI systems need to reason about. Customer history, transaction records, organizational structure, institutional knowledge accumulated over years of use — this is irreplaceable. The workflow layer sitting on top of that data is not.
For businesses currently evaluating their software spend, the calculation is changing. The right question is no longer "which SaaS product does this function best?" It's "which SaaS products store data that is genuinely irreplaceable, and which are just charging subscription fees to run workflows that Claude can handle?"
That assessment, done honestly, will look different for every organization. But the companies running it now — before pricing pressures force the conversation — will make better decisions than the ones that wait for a CFO mandate to cut SaaS spend.
Wrigley cancelled 7 subscriptions over a weekend. He's an early signal, not an edge case. The question is how many finance teams will run the same analysis before the NRR data shows up in quarterly reports and the market figures it out.