The “SaaSpocalypse” Isn’t What Wall Street Thinks It Is. I believe it’s Bigger.

I’ve been watching $300 billion in software market cap evaporate this week and I keep coming back to different conversations I’ve had with people building some of this tech. We were talking about what happens when the intelligence layer – Claude, GPT, Gemini – starts absorbing the application layer. Not competing with it. Absorbing it. One of the researchers said something that stuck with me: “The question isn’t whether the software survives. It’s whether it survives as a product or gets reduced to a feature.”

That’s the question Wall Street should be asking.

Jefferies is calling this the “SaaSpocalypse.” Traders are in “get me out” mode. Salesforce is down 26% on the year. Intuit is down 34%. Thomson Reuters dropped 16% in a single day – triggered by an Anthropic legal plugin. A legal plugin. Let that sink in!

And almost everyone is framing this as: “Is software dead?”

No. Obviously not. That’s a simple question that lets people avoid the harder one: which software companies have moats that survive when AI agents – not humans – become the primary user of the software?

If you think you have an advantage today because you built a nice UI on top of a workflow – rethink that thought. If you think your per-seat pricing model is safe because enterprises are locked in – rethink that also. If you believe the walls you’ve built around your customer relationships keep you protected – well, you may want to rethink that too.

Here’s why I am saying this.

The entire SaaS revenue model was built on headcount expansion. More employees, more seats, more revenue. AI compresses headcount. If 10 AI agents can do the work of 100 sales reps, you don’t need 100 Salesforce seats. You need 10. That’s a 90% reduction in seat revenue for the same output. Not hypothetically – operationally.

This morning Ulrike Hoffmann-Burchardi from UBS published a note this week that nailed the structural point most people are missing. Since ChatGPT launched: Nasdaq 100 up 115%. The SaaS index? Up 15%. That’s not a blip – that’s a three-year divergence reflecting a fundamental capital reallocation away from application-layer companies toward intelligence and infrastructure plays.

Jensen Huang from NVIDIA said this week that the idea AI will replace software companies is “the most illogical thing in the world.” His analogy: would an AGI invent a new screwdriver, or just use the one that exists? It’s a clever sound bite. But investor Gokul Rajaram got the rebuttal exactly right: if AI agents choose the tools, that’s commoditization. The agent doesn’t care about your brand, your sales team, or your golf outings with the CIO. It picks the optimal tool in real-time. That destroys pricing power even if the software survives.

Now here’s the part that I think the many from the broader tech and financial community may be underestimating: healthcare. And not for the reasons most people assume.

OpenAI’s January 2026 report revealed that 40 million people globally ask ChatGPT healthcare questions daily – with 70% of those conversations happening outside clinical hours. They’ve launched a dedicated healthcare suite. Anthropic just made its own healthcare move – not “Claude gives better medical answers” but a workflow and connector strategy: CMS coverage policy, ICD-10, NPI registry, PubMed – plus skills aimed at prior auth and interoperability. OpenAI is going bottom-up via consumer engagement. Anthropic is going sideways into admin and life sciences ops. Different entry points, same endgame: the agent layer is trying to become the system of engagement while the EHR stays the system of record.

I’ve been writing about this for a while now, and the framework I keep coming back to applies well beyond healthcare: System of Record != System of Interaction != System of Memory. The Record is the regulated ledger – EHR, CRM, ERP – it’s tied to billing, compliance, and legal discovery and it isn’t going away quickly. The Interaction layer is what humans (and now agents) actually use – an AI layer that sits above systems and turns intent into action. The Memory layer is the governed context spanning sources – interoperability, provenance, permissions, receipts.

That last one is where I believe lies the real battleground. Not just in healthcare – across every vertical getting hit in this selloff. The winner won’t be “best chat.” That’s table stakes. The winner will control the memory supply chain – the governed context layer that agents depend on to make decisions. And the moat isn’t conversation quality. It’s the judgment boundaries: when the agent stops, who it escalates to, what gets documented, and who owns the miss. That’s what survives. Everything else is a feature.

The existing software in healthcare is genuinely terrible. And the regulatory moats people assume protect incumbents? Thinner than most investors think. This is where the application-layer disruption will be most acute. Both OpenAI and Anthropic are now proving it in real time.

The Klarna experiment is the most honest data point we have outside healthcare. Their CEO spent a year trying to replace Salesforce. His conclusion: most companies won’t do what we did. But fewer SaaS companies will consolidate the market. That’s the real outcome – not death, but consolidation and brutal margin compression.

A CNBC anchor rebuilt Monday.com in an hour with Claude Cowork. That’s a fun demo. But it’s not the real threat. The real threat is that non-technical operators inside enterprises are building internal tools that eliminate the need to buy external SaaS. A YC company reported that a non-technical prospect showed up with sales workflows he built in Replit – replacing their paid SaaS tool entirely. That pattern should terrify every mid-market software company whose moat was “it’s easier to buy than build.” AI is quickly destroying that moat.

So what actually survives? Companies that own the data and context agents depend on. Infrastructure that agents run on. The governed memory layer – provenance, audit trails, accountability chains. Cybersecurity – because more agents mean a bigger attack surface, not a smaller one.

What doesn’t survive? Horizontal tools whose moat was a nice UI. Per-seat pricing without a transition path. Any application that can be replicated as a plugin. And any incumbent that thinks “system of record” protects them when the operating surface is moving to the interaction and memory layers faster than they can adapt.

And here’s the second-order risk from an investor perspective: Business Development Companies (think Ares Capital, Blackstone Secured Lending, Blue Owl Capital) hold roughly 20% exposure to SaaS through private credit. UBS models 13% default rates under aggressive AI disruption. The mid-market, per-seat SaaS companies in those portfolios are the most exposed names in the market.

Bank of America called this selloff “internally inconsistent” and compared it to the DeepSeek AI panic. They may be right that it’s overdone in the short term. But comparing this to DeepSeek misses the point entirely. DeepSeek was about the cost of building AI. This is about what AI does to every business model it touches.

The question isn’t whether enterprises will spend on software. It’s whether they’ll spend on your software – or redirect that budget to AI.

Rethink your tech stack. Rethink your moat. Rethink everything.