AI Didn't Kill SaaS. It Exposed What SaaS Actually Was.

There’s a comment buried 14 levels deep in this Hacker News thread about AI killing B2B SaaS. It has 37 upvotes and it’s the smartest thing I’ve read this year.

Here it is, paraphrased: “The real innovation of SaaS was laundering inaccessible open-source software into a format that doesn’t require transiting git. The hard part was never the code. The hard part was that git sucks.”

I laughed. Then I stopped laughing because it’s devastatingly correct.

---

Hey, I’m Lakshmi — I help developers build, deploy, and distribute their SaaS without hiring a team. I also run Stacksweller and Supabyoi.

New here? Start with Why Your AI Wakes Up Every Morning With No Memory or Clean Code Is Dead.

---

The Git Laundering Machine

Think about the most profitable SaaS businesses in technology. Seriously, list them.

AWS? That’s Linux, KVM, and Xen behind a billing dashboard. Heroku was git-push-to-deploy because deploying was too hard. Vercel is the same thing for Next.js. MongoDB Atlas is MongoDB without the ops. Redis Cloud is Redis without the YAML. Supabase is Postgres without the DBA.

Every single one of them is a factory that converts something freely available on GitHub into something you can pay for on a website.

The commenter was right. These companies didn’t build moats with proprietary technology. They built moats by standing between users and git. Their value proposition, stripped to the studs, is: “You don’t have to clone a repo.”

That’s a $500 billion industry built on the fact that git clone is scary.

LLMs Just Killed the Middleman

Here’s where the “AI is killing SaaS” thesis gets real.

When a CTO says “can we build this internally?”, the old answer was: “Technically yes, but you’d need 3 engineers, 6 months, and ongoing maintenance. Just buy the SaaS.”

The new answer: “ChatGPT set it up in 20 minutes. It reads from the same open-source code the SaaS vendor uses. It runs on our infrastructure. There’s no monthly bill.”

LLMs do exactly what SaaS companies do — they take inaccessible open-source software and make it usable by normal humans. They just skip the subscription.

The git laundering machine now has competition. And the competitor works for free.

What Actually Survives

So is B2B SaaS dead? No. But the moat map just got redrawn.

Here’s what doesn’t survive: any SaaS whose primary value is “we set it up so you don’t have to.” Deployment wrappers, config GUIs, managed hosting for commodity databases — all of this is getting compressed.

An HN commenter who manages teams put it bluntly: “Management doesn’t want to be responsible for bespoke internal tools.” That’s real. But it’s a shrinking moat. Today’s management doesn’t want to be responsible. Tomorrow’s management grew up with ChatGPT and doesn’t see internal tooling as risky.

Here’s what survives:

Data. If your SaaS accumulates proprietary data over time — customer behavior patterns, industry benchmarks, network effects — that’s a moat AI can’t replicate. A new LLM-generated tool starts with zero data. Your SaaS has three years of it.

Compliance and trust. SOC 2, HIPAA, GDPR certification takes time and money. “ChatGPT built it” doesn’t pass an enterprise security audit. Yet.

Workflow lock-in. Not the software itself, but the habits. Slack isn’t hard to replace technically. It’s hard to replace because your whole company’s muscle memory lives there.

Network effects. Figma isn’t valuable because of the rendering engine. It’s valuable because your designers, developers, and product managers are all in the same file. That’s a moat no amount of vibe coding can replicate.

The specification itself. Here’s the contrarian take within the contrarian take: as code becomes commodity, the spec becomes the product. The companies that survive aren’t the ones that write the best code. They’re the ones that understand the problem deeply enough to specify what “right” looks like. Everyone else is just a GPT wrapper with a landing page.

The Indie SaaS Playbook Changes

If you’re building SaaS solo — and if you’re reading this newsletter, you probably are — the implications are brutal and clear.

Full disclosure: I built a product that does exactly this. Supabyoi deploys Supabase for you. By my own thesis, that’s a shrinking moat. I’m writing this post partly because I’m living the question: evolve or get compressed.

Stop building tools. Start building data flywheels.

A CRUD app with a nice UI is now a weekend project for anyone with ChatGPT. A system that gets smarter with every user interaction is still a real business.

Stop selling setup. Start selling ongoing value.

“We deploy Postgres for you” is dying. “We analyze your Postgres performance patterns across 10,000 databases and tell you what’s about to break” is thriving.

Stop competing on features. Start competing on understanding.

The SaaS products that survive AI commodification will be the ones that understand their customers’ problems better than a general-purpose LLM ever could. Domain expertise is the last moat.

The $500 Billion Question

The HN thread devolved into the usual “AI is overhyped” vs. “AI changes everything” tribal warfare. But that one comment, buried 14 levels deep, cut through all of it.

The SaaS moat was never the software. It was the fact that software was hard to access. That moat is evaporating.

What’s left is data, trust, network effects, and deep domain understanding.

Build your SaaS around those. Or enjoy competing with a free chatbot.