back to Jitbit Blog home About this blog

AI is a feature, not a product

by Alex Yumashev · Apr 11 2026

About a year ago I wrote a piece called "Will AI destroy B2B SaaS?". In it I basically cried doom over the whole industry. "Build vs. buy has flipped", "LLMs default to the big names", "the SaaS graveyard is about to get crowded", I give it 4 years etc.

I was wrong. Well, half-wrong. Let me explain.

TL;DR

AI won't kill SaaS because building even an internal CRUD tool still requires an analytical mindset, clear thinking, systematic reasoning, structured intuition, conceptual clarity and a methodical perspective.

And most people are stupid.

I don't mean this as an insult. I mean it in the boring, statistical, half-the-population-is-below-median sense. I mean it as a guy who has been shipping B2B software for 20 years and has watched thousands of smart, capable, well-paid professionals fail to write a three-line Excel formula.

The "just vibe-code it" fantasy

The entire "AI will kill SaaS" argument goes like this:

"Why pay $29/month for a form builder when Claude can spit one out in 10 minutes?"

Yes. Claude can. I can. You can, dear reader, because you're reading a blog about code. But Linda from accounting cannot. Linda's boss cannot. Linda's boss's boss - the VP of Operations who signs the PO for that $29/month - definitely cannot. And not because they're dumb. They're smart in their domain. They just don't think in terms of "inputs, outputs, edge cases, null states, auth, data model, state management".

Because guess what - programming isn't about coding. Coding was never the bottleneck. Writing a five-table SaaS app on top of a database is 10% syntax and 90% figuring out what the damn thing should actually do. That part does not go away when Claude writes the code for you. Somebody still has to sit in the chair and say "actually, the invoice status should depend on the payment status, not the other way around, and also refunded invoices are a special case, and also we have partial refunds, and also the audit log shouldn't be editable, and also..."

Try walking an average office worker through that conversation.

Here's the whole thesis in one sentence, so nobody misses it: building an internal tool - even a dumb, ugly, 4-table CRUD app - requires an analytical mindset that most people simply do not have. AI or no AI. Claude or no Claude. You can hand someone the best code generator on Earth and they will still fail at the part before the code.

"But enterprises will build in-house"

Ha. No they won't.

Enterprises have been "building in-house" since the dawn of time. That's why every Fortune 500 has a graveyard of abandoned SharePoint sites, half-finished Access databases from 2009, and a Jira instance that nobody owns anymore. Building software inside a big company is a political minefield, not a technical one. Who owns it? Who pays for it? Who's on-call at 3am when it breaks? Who updates it when the tax law changes? Whose budget is it in next year when the director who championed it gets reorged?

This is precisely why buying SaaS exists. You're not paying Jitbit $29 per agent because we wrote some clever code. You're paying us so that when your helpdesk breaks on a Tuesday morning at 9am, it's our problem, not yours. That bit - the responsibility transfer - doesn't get cheaper when AI shows up. If anything it gets more expensive, because now you also need someone to babysit the hallucinating agent.

Steve Jobs and Dropbox

Here's the part I didn't want to admit a year ago.

For a while I was convinced the AI providers themselves - OpenAI, Anthropic, Google - would become the new behemoths. That they'd "Amazon-Basics" the whole software industry, casually absorbing every category underneath their umbrella. Write emails? We do that. Summarize docs? We do that. Run your helpdesk? Yep, we do that too. The big labs would eat the app layer for breakfast and the rest of us would be left selling consulting hours to whoever's still standing.

I don't think that anymore. I think GPT (and Claude, and Gemini) are going to follow the Dropbox playbook instead - and that is actually a much more interesting story.

When Steve Jobs was trying to acquire Dropbox back in 2009, Drew Houston turned him down. Jobs - famously - said: "Dropbox is a feature, not a product." Everyone laughed at the time. Dropbox went on to IPO at $10B. Jobs looked silly.

Except... he was right. He was just early.

Today, who actually uses Dropbox? A tiny, shrinking subset of people. The "sync your files to the cloud" thing is now a feature inside iCloud, inside Google Drive, inside OneDrive, inside every operating system on Earth. Dropbox the company still exists, still makes money, but it's a fraction of the ambition. The feature ate the product. Jobs was right. It just took 15 years.

AI is the same story.

"AI" is not a product. "AI writing assistant" is not a product. "AI customer support bot" is not a product. They're features. Features that will live inside Word, inside Gmail, inside Zendesk, inside Jitbit Helpdesk, inside every SaaS app you already pay for. The standalone "AI something" startups that raised $50M at a $500M valuation to build "Cursor for X" - those are the Dropboxes of 2026. Every Cursor needs VSCode engine to build on top of.

This is also why I don't worry too much about AI replacing Jitbit. It will be a brilliant feature inside our product. It already is - we ship AI ticket summaries, AI suggested replies, AI categorization, an MSP-server for external agents... Our customers love it. They didn't cancel us and go build their own helpdesk. They just started clicking the "summarize" button.

So what did I get wrong a year ago?

A year ago I confused "AI can generate code" with "AI can generate products". Those are wildly different things.

Code is the easy part. Always has been. The hard part is:

AI helps with exactly zero of those. Maybe 0.5 of them.

Who should actually be scared

To be fair, my original post wasn't entirely wrong. Some SaaS will die. Specifically:

  1. One-trick pony tools. Form builders, QR code generators, "prettify my JSON" utilities. These were already dying. AI just speeds up the funeral.
  2. Thin wrappers around OpenAI. You know who you are. "AI-powered email writer" = system prompt + GPT-4 + Stripe. That's not a business, that's a weekend project with a landing page.
  3. Anyone whose entire moat was "we wrote the code". If the code is the moat, the moat just got a drawbridge.

But "real" SaaS - the kind with a data model that took a decade to get right, with integrations that took five years to build, with a 99.95% SLA and a support team and a brand people trust - that's fine. Better than fine, actually. It's going to get more valuable, because everyone's going to be drowning in AI-generated half-broken internal tools, and someone will have to rescue them.

Which brings me back to a line from last year's post that I'll happily leave in:

"Software consulting is making a comeback though. Someone has to clean up the vibe-coded chaos."

Yep. That part I stand by.