Is SaaS Dead? It's Complicated...

Welcome back.

Software stocks just had their worst week in years. $285 billion evaporated. Traders coined a new term: "SaaSpocalypse."

Everyone has an opinion, and most of them are missing the point.

So I wanted to take a look at what actually happens when foundation models move into the application layer, explore why "taste" might be the only moat that matters, and what this means if you're building (or buying) software.

Let's dig in.

Over a year ago, Satya Nadella declared on the BG2 podcast that "SaaS is dead."

At the time, many dismissed it as hyperbole. Then this Monday happened.

$285 billion evaporated from software stocks after Anthropic announced new Claude Cowork plugins that automate tasks like legal document review and compliance workflows. Goldman Sachs' software basket dropped 6% in a single day.

I've spent the past few days reading every take on this. Here's my honest assessment: the doomers are wrong, but so are the dismissers.

Nadella's argument is simple. Why click through Salesforce when an AI agent can just do the work for you?

Nicolas Bustamante put it sharply in his Substack post: "The interface moat is dead. The data moat remains. But for most SaaS companies, the interface was the moat."

That last line stings because it's true. Most B2B software isn't valuable because of proprietary data. It's valuable because someone built a decent UI for a workflow.

If Claude can build that UI on the fly, what are you paying for?

Steven Sinofsky, former Windows President, wrote a lengthy rebuttal titled "Death of Software. Nah."

His argument: we've seen this movie before. PCs were supposed to kill mainframes. The web was supposed to kill desktop software. Mobile was supposed to kill the web. None of it happened.

He thinks we're looking at 20-30 years of gradual transition, not a cliff.

Aaron Levie, Box's CEO, made a similar point: "Try vibe-coding an ERP system that handles payroll for 10,000 employees across 40 countries. Good luck."

Enterprise software is boring precisely because it handles edge cases that took decades to discover.

Jensen Huang piled on Tuesday, calling the selloff "the most illogical thing in the world." His argument: AI uses tools, it doesn't replace them.

Matthew Berman proposed the question: Why pay for software when you can just build it yourself with AI? He's literally going through his SaaS stack, rank-ordering by complexity, and rebuilding each app from scratch.

Here's where I break from Sinofsky.

The previous disruptions he mentions (PCs, web, mobile) were platform shifts. You still needed software engineers to build on each new platform. The barrier to entry stayed high.

AI agents are different. They lower the barrier to building software to nearly zero.

I've already replaced my journaling app with a vibe-coded alternative. Same with my read-later app. They're not as polished, but they're exactly what I need.

A year ago, I couldn't have done that. Now it takes an afternoon.

Here's the thing: I still pay for Obsidian. I still pay for Evernote. I still use Slack.

Why? Because taste is a durable moat.

Great software anticipates needs I didn't know I had. It makes decisions I don't want to think about. It has opinions.

There's another factor too: organizations aren't bottlenecked by code. They're bottlenecked by getting humans to agree on what to build.

Vibe-coding doesn't solve the consensus problem. It just makes the coding part faster.

I'd call it Software Darwinism, not Software Apocalypse.

Commodity software (the stuff that's just a thin UI over a database) is in trouble. The $285 billion selloff is pricing that in, probably too aggressively. J.P. Morgan's Toby Ogg put it well: the sector "isn't just guilty until proven innocent but is now being sentenced before trial."

But software with genuine taste, network effects, or deep enterprise integration? That's not going anywhere.

The pie isn't shrinking. The slices are getting rearranged.

Is SaaS dead? No.

Is the SaaS business model under pressure? Absolutely.

The companies that survive will be the ones that are more than just a UI. They'll be the ones with taste, with network effects, with data moats, or with the kind of boring enterprise complexity that no one wants to vibe-code from scratch.

If you're building software, ask yourself: Would a motivated person with Claude replace me in an afternoon?

If the answer is yes, you have a problem.

I'm genuinely curious about this one.

This week's question: Look at your company's SaaS stack (or your personal one). Which tools are you now reconsidering because AI could do the job? What's stopping you from building your own version?

Hit reply with your thoughts. Best responses get featured on Tuesday.

Until next week,

Haroon