The SaaSpocalypse and the questions nobody's asking

This week, the software sector experienced one of its most violent selloffs in years.

Over $300 billion in market value evaporated in two days. Thomson Reuters down 18%. RELX down 14%. LegalZoom down 20%. Enterprise giants like Salesforce and ServiceNow down 7%. The selling was indiscriminate and brutal.

The catalyst was clear: Anthropic released plugins for Claude Cowork that moved from "AI assists humans" to "AI executes labour autonomously." Legal work, financial analysis, data tasks, sales and marketing — all positioned as deliverable with minimal human intervention.

Markets interpreted this as existential. One trader called it "get me out at any price" selling.

But as the dust settles, the more interesting question isn't what happened. It's what the market is now pricing in — and whether that assumption holds up under scrutiny.

What the market is assuming

The breadth and speed of the selloff reveal an implicit belief: that most software, as it exists today, will become insourced business functions tomorrow.

The car workshop won't just replace its accounting software with Claude. They'll replace their ERP, their electronic parts catalogue, their booking management system, and their CRM — all with AI-assisted custom tools they build and maintain themselves.

Software-as-a-Service becomes Software-as-DIY.

This assumption explains why the selling was so indiscriminate. If foundation models can automate workflows at consumer pricing, why pay enterprise software companies hundreds of dollars per seat per month?

It's a compelling story. And it contains real truth.

But it also contains a very large, very convenient gap.

The questions nobody's asking

Let's think through what universal AI-driven insourcing would actually require:

1. Do businesses want the responsibility?

A car workshop might be happy to vibe-code a basic booking tool. But do they really want to oversee the ongoing maintenance, improvement, and integration of six to ten different business-critical applications?

At some point, managing custom software stops being a cost-saving and becomes a distraction from the actual business — fixing cars.

2. Where does support go when something breaks?

Software companies provide 24/7 support, guaranteed uptime, and contractual SLAs. When a mission-critical tool fails at 2 AM, there's someone to call.

If a business builds its own tools with Claude, who do they call when the AI-generated code breaks in production? Anthropic doesn't offer enterprise support for vibe-coded applications. Neither does any other foundation model company.

3. What about regulation and compliance?

Some software exists in heavily regulated environments: healthcare, financial services, legal discovery, tax compliance. These aren't workflows you can automate casually.

Will regulators accept AI-generated tools with no audit trail, no version control, and no accountability when things go wrong?

Probably not.

4. How does ongoing improvement work?

Enterprise software companies employ teams of engineers who continuously improve products, patch security vulnerabilities, and adapt to regulatory changes.

If a business insources its tools, that engineering responsibility doesn't disappear — it just shifts to them. And most businesses don't have the capability, or desire, to become software companies.

5. Not all software is created equal

Here's where the indiscriminate selling reveals its flaw.

Some software is genuinely simple: basic kanban boards, lightweight CRMs, templated document generation. These tools are vulnerable to AI replacement because they're essentially data entry plus display logic.

But other software is deeply complex: mission-critical ERPs with decades of embedded business logic, regulatory compliance platforms that require constant revision, systems with intricate integrations across dozens of other tools.

The market sold both categories equally. That doesn't make sense.

What durability looks like

The businesses most at risk are those whose value proposition was always "we automate a simple task better than you could yourself."

If the task is genuinely simple, AI can do it. And the business had no moat to begin with.

But businesses with genuine durability have different characteristics:

Deep regulatory integration. Software that must comply with constantly evolving legal or industry standards requires institutional knowledge and ongoing maintenance. AI can assist, but it can't own the accountability.

Mission-critical complexity. ERPs, tax software, healthcare systems — these aren't amenable to vibe coding. When failure means regulatory penalties or operational collapse, businesses don't experiment.

Network effects and data moats. If the software's value comes from aggregating data across users (think freight logistics platforms or industry benchmarking tools), AI can't replicate that by running locally.

Irreplaceable integrations. Software that connects dozens of other systems and maintains those integrations as they evolve provides ongoing value that one-off AI tools can't match.

The market is pricing all software as if it's in the first category. But the businesses in the second category just became dramatically cheaper.

Where the market might be wrong

We think the market has swung toward bareknuckle optimism about limitless AI entrenchment — without seriously considering the practical constraints.

Yes, AI will replace some software. Probably a lot of it.

But the assumption that businesses will happily insource all their tooling ignores basic realities:

Most businesses don't want to be software companies. They have a core competency, and it's not managing custom applications.

Foundation models are incredibly powerful, but they're not enterprise support infrastructure. They're not compliance frameworks. They're not institutional knowledge.

And perhaps most importantly: the failure modes are different.

When enterprise software breaks, you have a vendor to hold accountable, contractual remedies, and insurance. When your AI-generated tool breaks, you have... a prompt you can't debug and no one to call.

What we're watching

This selloff has created a sorting problem — and potentially an opportunity.

Software companies with genuine embedded value, regulatory moats, or mission-critical complexity are now trading at distressed valuations alongside commoditised workflow tools.

Over the coming months, we'll be watching:

• Which companies can demonstrate genuine pricing power and customer retention despite AI alternatives
• How enterprises actually adopt (or don't adopt) AI-generated tools for critical functions
• Where regulation draws boundaries around AI automation
• Which software businesses had durable advantages all along — and are now on sale

Bringing it back to durability

The right question isn't "Will AI replace software?"

It's "Which software businesses were built on genuinely durable foundations, and which were renting workflows they never truly owned?"

The market is currently pricing both the same. We don't think that lasts.

Inside Baseline Capital, we're working through this question company by company — stress-testing assumptions, mapping failure modes, and distinguishing between software that automates simple tasks and software that owns institutional complexity.

The panic creates noise. The noise creates mispricing. And mispricing creates opportunity.

More on that soon.

— Baseline Capital

[Request access]

━━━━━━━━━━━━━━━━━━━━━━━━━━

END TRANSMISSION