The End of Selling Software to Technical Users

There's a stat from Retool's latest report that I keep coming back to. 35% of teams have already replaced at least one SaaS tool with something they built themselves, not just considered it. And 78% expect to build more custom tools this year.

And it's only going in one direction.

For decades, there's been a massive economy around selling tools to technical users. Dev tools, CLI utilities, internal dashboards, data pipelines, monitoring systems, project management, analytics platforms. The pitch was always the same: yes, you could build this yourself, but it would take months and you have better things to do. So you pay $20 per seat per month and move on.

That math made sense when building was expensive, but the cost of building has collapsed.

I've watched this shift happen in my own work. I used to pay for tools that solved problems I could describe in a paragraph. A dashboard that shows me X. A script that does Y on a schedule. A workflow that connects A to B. Now when I run into one of those problems, my first instinct isn't to search for a SaaS product. It's to open Claude Code and build exactly what I need in an afternoon. And I don't mean a half-working prototype. A real, working tool that does precisely what I want, nothing more, nothing less.

And I'm not even an engineer. I'm a product manager who codes. If I can do this, the actual engineers are way ahead of me.

The numbers back this up. A quarter of startups in Y Combinator's latest batch had codebases that were 95% AI-generated. Google says 25% of their code is now written by AI. Microsoft says 30%. GitHub reports that nearly half of all code written on their platform in 2025 was AI-generated. The cost of building software has collapsed so fast that the entire "build vs. buy" calculation has flipped.

What's Getting Eaten First

Internal admin tools and dashboards. BI and analytics platforms. CRMs. Form builders. Workflow automation. Project management. Content tools. These are, as Satya Nadella put it, "essentially CRUD databases with a bunch of business logic." That's the CEO of the largest enterprise software company in the world saying the entire application layer is about to collapse into AI agents. When the person selling the software is telling you the software is going away, you should probably listen.

The SaaSpocalypse

The Wall Street version of this story already has a name. They're calling it the "SaaSpocalypse." A leaked memo from a Fortune 50 company about cutting Salesforce and ServiceNow licenses by 60% helped trigger a trillion-dollar wipeout in SaaS market cap earlier this year. Salesforce stock dropped 30%. Investors are pricing this in before most SaaS companies have even acknowledged it's happening.

And it's not just that people are building custom replacements. It's that the open source ecosystem is about to flood with alternatives. When the marginal cost of building a tool approaches zero, you don't just get more custom internal tools. You get thousands of people building and sharing the tools they made. Every niche utility, every opinionated workflow, every domain-specific dashboard that used to justify a $50/month subscription is going to have three free alternatives on GitHub by the end of the year. They might not be polished, but they'll be good enough. And for technical users, "good enough and free" beats "polished and $20/seat" almost every time.

Where This Argument Breaks

I want to be honest about where I think this argument breaks, though.

AI-generated code has real problems. Fixing bugs in AI-generated code costs 3-4x more than in human-written code because of the context gap. The code works, but the person who prompted it often doesn't fully understand why. One study found that code churn, rewrites of recently written code, has doubled since AI coding tools became mainstream. Another found that experienced developers were actually 19% slower with AI tools, despite believing they were 20% faster. The "vibe coding hangover" is real. A lot of teams that enthusiastically replaced their SaaS tools with custom builds in 2025 are now discovering what it means to maintain code nobody fully understands.

And there are categories of software that aren't going anywhere. Payment processing, anything requiring five nines of uptime, heavily regulated systems. You're not going to vibe-code your way to a Stripe competitor. Platforms with deep network effects, like Slack, where the value comes from everyone else being on the same platform, are safe too. Same for software that encodes years of domain expertise and compliance logic. The hard part isn't the interface, it's the thousands of business rules underneath it that took a decade to get right.

Infrastructure Survives, Applications Don't

But here's the thing. That's a much smaller market than what exists today. The stuff that survives is infrastructure, not applications. Stripe survives. The fifteen analytics dashboards your company pays for probably don't. Snowflake survives. The BI tool sitting on top of it, the one that just translates your questions into SQL and makes charts, probably doesn't. The platform survives. The application layer on top of it gets rebuilt by every team for their own needs.

I think the companies most at risk are the ones that built a business on the gap between "I know what I want" and "I can build what I want." That gap used to be wide. It used to take months of engineering time to cross it. Now it takes an afternoon. And every month, it gets shorter.

Tedious vs. Hard

If you're building a SaaS product for technical users, I'd be asking myself a hard question: is the thing I'm selling genuinely difficult to build, or is it just tedious? Because AI is about to eliminate tedious, but it's nowhere close to eliminating hard. The companies that are selling hard will be fine. The ones that are selling tedious on a subscription are in trouble.

I don't know exactly how this plays out. Maybe the SaaS companies adapt and become AI-native fast enough. Maybe the maintenance costs of custom-built tools push people back toward buying. Maybe something I'm not seeing changes the math entirely. But the direction is clear, and the data is already there. A third of teams have replaced a SaaS tool. That number is only going one direction.

The era of paying $20/month for something a technical user can build in a day is ending. The question for a lot of companies isn't whether it happens. It's whether they see it coming in time to build something that's actually hard to replace.