The customer always comes back
A friend of ours runs a successful SaaS business. She was giving us an update on the latest AI trends she’s seeing, and our recent conversation make me laugh, but then think (which is unusual for me).
The boogeyman of AI for every SaaS business right now is the idea that customers will cancel their subscriptions to your tool because they vibe-coded their own version of the product. Dashboard, charts, the whole thing, built it in a weekend with Cursor or Claude. And this is happening to day, they build a demo in a weekend and it looks awesome.
This has happened to our friend over a dozen times already, the team panicked… but then 2-3 months later, the customers started coming back, and this happened for each of the dozen customers.
The demo is not the product
We’re living through an incredible moment for building software. A non-technical person can sit down with an AI coding tool and produce something that genuinely looks like a real product. This wasn’t possible a year ago, and we shouldn’t downplay it. As you might imagine, I’m about as far as you can get from “this AI stuff doesn’t work.”
Still, there’s a gap between “looks like it works” and “actually works.”
Our friend experienced this firsthand looking to deploy features to her product using AI. It connected to real APIs, pulled real data, had a clean UI. The problem? Some months, the math was just wrong. The app made numbers up.
She was up at 2 a.m. screaming in all caps at Cursor before she gave up and decided that maybe the time had come to consider the AI effort the demo and proof of concept and actually build the feature from the ground up “the right way.”
I’m not knocking her or the tools, it’s just the reality of software. The last 20% of the work is 80% of the effort. Getting data to display is easy, getting it to display correctly, across edge cases, over time, with real users, well, that’s the real value a product provides.
The $1,000 problem
There’s a meme going around about someone complaining that $1,000 a month for HubSpot is too expensive. The punchline: they replace it with a full-time hire at $100,000 a year to build a worse version.
The math on building your own tools almost never works out. Our friend’s AI feature prototype took about 10 hours to build something that sort of worked. If they actually wanted to deploy that feature, meaning that we’re taking into consideration reliable data, period-over-period accuracy, maintainability, etc, we’re still days if not weeks out.
For this specific feature she estimated a 15 month break even on the time investment compared to literally just doing the task manually. Your mileage will vary depending on the feature you’re building, how frequent and time-consuming the task is, and how close you get to a “one shot”, but everyone I’ve spoken to starts to think that maybe they should just keep doing the task “by hand” after three months of using their vibe coded tooling.
I get the annoyance around subscriptions, but in my (ultimately biased perspective) it’s still an absurdly good deal compared to every alternative. A software product is hundreds or thousands of engineering hours amortized across a customer base. When you build your own, you’re paying full price for all those hours yourself. You get infinite customization and flexibility, and there are situations where it’s too painful to bend the existing product around your org, but just under the surface are all those implementation details, lying in wait.
The “hold the line” moment
I had a call with the folks at another OG SaaS company and they mentioned something interesting: their bigger customers all want to do “agent” stuff in-house right now. Instead of paying the company that’s building the product to do the “agent layer”, they want to take the data returned by the product and build it themselves.
This is happening everywhere, Enterprise teams feel empowered by AI tooling (and motivated by the relentless enterprise sales teams at all these companies) and want to try. That’s healthy, but it’s also a phase.
Right now the momentum is on the side of stories like “I vibe-coded Salesforce in a weekend.” That narrative is exciting and new and gets engagement on LinkedIn.
My prediction is that soon the momentum will shift to “the customer came back.”
Teams will (and should) do their limit testing. They’ll discover that connecting to an API is the easy part, and maintaining accuracy, handling edge cases, and iterating on UX over months is the actual job. And then they’ll be ready to work with purpose-built tools again, with a much better appreciation for what those tools actually do.
Ever since “fours steps to the epiphany” we’ve all known that the best customer recognizes they have a problem, has spent time and money to try and solve the problem, has failed at solving the problem, and now they know exactly the solution they need (and how hard the problem actually was.) AI speedruns potential customers through this process at lightspeed, we just need to hold the line as the limit testing occurs.
What SaaS companies should do right now
If you’re a SaaS company watching customers churn to build their own: don’t panic. And definitely don’t slash your prices or start bolting on AI features just to have a story to tell.
Instead do the work that AI-generated tools can’t replicate: deepen your domain expertise, tighten your feedback loops with customers, and make your product the thing that’s not worth rebuilding.
A non-technical person summed it up better than any analyst could: “I could make something that looks ok but there is zero chance I could get it across the finish line. I need help.” That gap between “looks ok” and “finish line” is where your product lives. It’s where your years of iteration, customer feedback, and domain knowledge compound into something that a weekend project can’t touch.
If ever there was a time to triple down on product, now is the time. The joke is that when your customers look up from their experimentation replacing internal tools they’ll have the same revelation: hey, maybe we should be using these AI tools to make OUR PRODUCT better for customers too?
If you want to make sure your customers actually know about the work you’re shipping while they’re out trying to rebuild your product, let’s have a chat about how Changebot keeps your audience in the loop automatically.
