The Enhancement Doctrine.

The Inciting Event: $1 Trillion Wiped in Weeks

Between February 3rd and 5th, 2026, approximately $285–300 billion in software market value evaporated in 48 hours. The IGV — the iShares Expanded Tech-Software ETF — entered a technical bear market, down roughly 30% from its September 2025 high. Across the sector, total estimated losses exceeded $1 trillion.

Salesforce dropped 26%. ServiceNow fell 23%. Adobe, Workday, and Zoom were all hit hard. These weren't companies with bad quarters. They were companies caught in a narrative collapse.

The trigger was Anthropic's launch of Claude Cowork — an agentic system perceived by the market as capable of replacing the knowledge workers who actually use SaaS tools. The logical conclusion wasn't subtle: if the agent does the work, why pay for the seat?

This wasn't panic without precedent. The market has overreacted to structural threats before — cloud, mobile, open source. But this time the mechanism was different. The threat wasn't to one company's market share. It was to the fundamental unit of commerce that the entire SaaS industry was built on.

The implied conclusion was almost algorithmic in its logic: an AI agent handling the work of ten customer service reps means a company loses 90% of the per-user license revenue from that deployment. The economics don't bend gradually. They collapse.

The Bear Case: Why the Panic Was Rational

Let me be honest about something: the bear case wasn't crazy. The structural threat to per-seat licensing is real. The Klarna story proved it.

The Death of the Seat

Klarna ditched Salesforce and Workday entirely, consolidated onto an AI-augmented stack, and deployed an OpenAI-powered system that handled the equivalent work of 700 employees. That's not a rounding error. That's a category-level displacement event. When agentic AI can do the work of ten customer service reps, the client loses 90% of per-user license revenue from that deployment. The economics collapse almost instantly.

Salesforce recognized this before the selloff — they were already pivoting to a consumption-based model, charging roughly $2 per action rather than per seat. That's not a minor adjustment. That's a fundamental business model disruption executed mid-flight. Credit to them for the pivot, but the market wasn't wrong to read it as an admission that the old model was broken.

The UI Moat Was Always Shallow

Databricks CEO Ali Ghodsi made the point as cleanly as anyone: when the interface becomes natural language, the moat built around a well-designed UI collapses. If the value of Salesforce or ServiceNow lived in the way their dashboards looked and worked, and a general-purpose model can now replicate that interface — then the moat was shallower than anyone admitted.

This hit close to home for a lot of horizontal SaaS companies. The ones that had been selling the workflow, not the data underneath it. The ones whose defensibility was the interface, not the proprietary information that flowed through it.

The Budget Displacement Problem

There's also a simpler, harder-to-argue-with dynamic: capital is finite. Meta is spending up to $135 billion on AI capex in 2026 alone. Hyperscalers collectively are deploying $660–690 billion on AI infrastructure. Every dollar going to AI tooling is a dollar not going to a new Salesforce seat, a Workday module, or a ServiceNow add-on. Enterprise budgets don't scale infinitely. The competition for them has gotten more intense.

The Bull Case: Why the Panic Was Overdone

Here's where I want to push back — hard — because the market made a classic mistake. It priced in a permanent worst-case scenario without accounting for the actual fundamentals underneath these businesses. More importantly, it ignored something critical: valuations had already regressed to pre-ZIRP norms. The selloff compressed multiples below what these companies earned even during the disciplined era before free money distorted everything.

The ZIRP Distortion Nobody Talked About

To understand how overdone the selloff was, you have to understand where these valuations came from in the first place. The SaaS market spent 2020 and 2021 pricing in a world where capital was essentially free, growth was valued above everything, and the terminal value of any subscription business with a good NRR was assumed to be infinite. That world was always fictitious.

Four company stories tell the real arc.

Adobe completed its pivot from perpetual licenses to Creative Cloud in 2018 and traded around $90 per share. By November 2021 — peak ZIRP — it hit $688. Today it sits near $273. That sounds like devastation until you realize: $273 is still triple the 2018 price. Adobe's business is materially larger and more profitable than it was when the market thought $90 was fair value. The selloff didn't price in business deterioration. It priced out multiple expansion.

Zoom is the purest expression of ZIRP distortion. At its peak in October 2020, the stock hit $588 per share — a $160 billion valuation — driven by pandemic-era demand that the market assumed was permanent. Today it's around $75, with a market cap of roughly $25 billion. Zoom's revenue kept growing after the pandemic peak. The company cut costs aggressively and is now more profitable than it was at $588. The stock is down 87% from its highs while the underlying business is fundamentally healthier.

ServiceNow peaked near $740 in November 2021 and trades around $100 today — an 85% decline from peak. And yet: 2026 subscription revenue guidance is $15.5 billion, up from $12.9 billion in 2025. The business is growing faster than the stock price suggests it should be. ServiceNow went from trading at roughly 120× sales at peak ZIRP to approximately 5× sales today.

Workday peaked near $340 in November 2021 and trades around $136 today — but that's still above its 2018 price, and the company now carries significantly more revenue, better margins, and real profitability. The less violent decline from peak compared to Zoom or ServiceNow is itself signal. The market is making a distinction between companies with real regulatory moats and switching costs and those whose value lived primarily in their interface.

The pattern across all four is consistent. 2026 valuations are tracking back to 2014–2019 norms: roughly 5–10× revenue for healthy recurring-revenue businesses. But the businesses themselves are larger, more profitable, and more efficient than they were a decade ago. The selloff compressed multiples below what these companies deserved even in the pre-ZIRP era. That's the overshoot. And that's what the relief rally corrected.

Jensen Huang's Counterargument

NVIDIA's CEO called the replacement narrative what it is: illogical. His argument isn't sentimental — it's structural. AI systems are massive consumers of software infrastructure. Operating systems. Databases. Developer tools. Security layers. Compliance frameworks. AI doesn't replace software; it runs on top of it. The more AI you deploy, the more software infrastructure you need to manage, govern, and run it.

Vertical SaaS Has Stronger Moats Than the Market Priced

Here's where I'd add a layer that I think the broader market has consistently underweighted: horizontal and vertical SaaS are not the same business, and they don't face the same threat.

A general-purpose AI agent can replicate a lot of things. It cannot replicate years of proprietary industry data, regulatory compliance infrastructure built for healthcare or finance or construction, or the workflow depth that comes from a decade of domain-specific customization. These are real moats. Not perfect moats — but real ones.

The companies that are most exposed to AI displacement are the ones whose value was always in their UI. The ones whose core defensibility is the data underneath the interface — the industry-specific logic, the compliance workflows, the accumulated transaction history — have a legitimate path forward. The market, in its panic, didn't make that distinction carefully enough.

The Earnings Said Something Different

While the narrative was collapsing, the actual business results weren't. ServiceNow's 2026 subscription revenue forecast came in at $15.5 billion, up from $12.9 billion the prior year. Salesforce beat revenue estimates in Q4 FY2026, with Agentforce and Data Cloud showing accelerating enterprise adoption and record remaining performance obligations. Global software spending is projected to grow 14.7% to $1.43 trillion in 2026. AI-related software spending is expected to triple over 2025–2026.

These are not the numbers of a dying industry. These are the numbers of an industry in transition.

The Turning Point: The Enhancement Doctrine

The relief rally didn't come from better earnings alone. It came from a narrative shift — the most important one the industry has seen in years.

On February 25th, 2026, Anthropic announced ten strategic enterprise partnerships. Partners' shares saw significant gains on opening. The market coined a phrase almost immediately: the Enhancement Doctrine. The thesis that AI and incumbent software don't compete — they coexist and amplify each other.

The reframe is important. AI models need enterprise data. They need workflow context. They need domain expertise. They need the compliance infrastructure and the proprietary datasets that only existing SaaS platforms hold. The Anthropic partnerships weren't an act of charity toward incumbent software companies. They were an acknowledgment that the most sophisticated AI labs in the world need enterprise software to function at scale.

This is the inverse of the Klarna story. Klarna circumvented the platform. The Enhancement Doctrine is AI augmenting the platform — using it as the data layer and workflow context that makes agents actually useful in enterprise environments.

Both things can be true simultaneously: some companies will get Klarna'd, and others will become indispensable infrastructure for AI systems. The determining factor is whether your value lives in your UI or in your data.

What This Means Going Forward

The New Playbook for Incumbent SaaS

The strategic imperative has shifted. Stop selling seats. Start selling outcomes. The question every SaaS company needs to answer isn't "how many users do you have?" — it's "what is the verifiable business outcome your product produces, and can you charge for that outcome rather than for access to the tool?"

The second imperative is to prove you are the data layer that AI needs to function in your vertical. Not the interface. The data. The integrations. The compliance infrastructure. The workflow context that a general-purpose model can't replicate from a cold start.

The companies that will win are the ones that position themselves as infrastructure AI runs on — not as interfaces humans use. Those are two fundamentally different value propositions, and the market is going to continue sorting them apart.

Roughly 45% of engineering and product resources protecting and deepening core functionality, roughly 45% building AI-native features within existing workflows, and a small skunkworks team — maybe 10% — building genuinely AI-native products that could eventually cannibalize the core. Companies that don't do the last part will have it done to them by someone else.

The New Investor Framework

The old framework — growth at all costs, TAM expansion, seat count, net revenue retention as the only metric that mattered — is over. It was always a product of free money. The new framework asks different questions: What is your capital efficiency? Where are your AI integration proof points? How deep is your data moat in your specific vertical? If you shift to consumption-based revenue, how durable is that revenue at scale?

Vertical SaaS at the right price is, I believe, one of the more compelling places to invest in the current environment. Deep moats at compressed multiples — companies whose value lives in proprietary data and regulatory positioning, priced as if they're already dead — is the opportunity the SaaSpocalypse created.

The Unresolved Question

I want to end honestly, because the bear case hasn't been fully refuted — it's been challenged. The bulls argue this is a cyclical panic, structurally similar to 2016 and 2022. Enterprise software has absorbed existential-looking threats before and compounded through them. The bears argue this time is architecturally different — that agentic systems represent a genuine displacement mechanism, not just competitive noise, and that the seat-based model's collapse isn't a temporary disruption but a permanent structural shift.

Both positions have merit. The market is effectively pricing in that the verdict will be delivered by 2026 earnings season. I think the truth will end up being more nuanced than either camp currently admits: some companies will be genuinely disrupted, some will become more valuable because of AI, and the ones caught in the middle — with shallow moats and interface-dependent value propositions — will struggle in ways that look increasingly permanent.