Why your SaaS needs tests: one missing 2-minute test let a single click approve 15 applications by mistake. The real cost of skipping tests — 112 hours.
Tests aren't overhead — they're insurance. I skipped them for 15 years, until one missing two-minute test let a single click approve 15 startup applications by mistake, in front of a paying customer. Here is what that bug actually cost, and why AI finally makes testing affordable for a solo founder.
I'll be honest about why I never wrote tests: they took longer than the feature, they were boring, deadlines always won, and "I'll add tests later" — narrator: he never did. Most solo builders do exactly the same. Until the day it stops being a theoretical risk.
The bug that approved 15 applications
March 2024. I had just launched the application review feature in my platform — the screen where VCs and accelerator managers review startup applications, score them, and make decisions. Critical feature, high stakes. I tested it manually. It worked perfectly.
Then a VC called me. "Luis, I just approved 15 applications by mistake. I clicked on one, and they all got approved."
My stomach dropped. I checked the code and found the bug immediately: a copy-paste error. Instead of approving the one application that was clicked, the function quietly fetched all applications and approved the lot. A rookie mistake — and no test to catch it.
The VC had to manually reverse 15 approvals. That one bug cost me three hours of emergency fixes, multiple apology emails, a dent in my credibility, and a customer who almost churned. And the test that would have caught it — create two applications, approve one, check that the other is still pending — would have taken two minutes to write.
Two minutes of prevention versus three hours of damage control, plus my reputation. That was the day the math stopped making sense.
The real cost of shipping without tests
After that bug, I audited the previous six months of building. The numbers were worse than the story:
| Six months without tests | The damage |
|---|---|
| Bugs shipped to production | 47 |
| Average time to discover each one | 3.2 days |
| Average time to fix each one | 2.4 hours |
| Total debugging time | 112 hours |
| Value of that time at €80/hour | €8,960 |
112 hours of my life spent debugging bugs I could have prevented. And the unmeasurable part: how many users quietly left because of those 47 bugs? Hard to know. Definitely not zero.
Meanwhile I kept telling myself I "didn't have time" to write tests. The truth: I had time to debug. I just didn't have time to prevent bugs. Read that sentence again — it's the lie that keeps most solo founders shipping broken software.
Why solo founders skip testing (and why it used to be rational)
Let's be fair to past me. Tests are repetitive and uncreative. Writing comprehensive tests for a two-hour feature could take another three hours — 60% of your time on work no user ever sees. Features feel like progress; tests feel like busywork. And there's always "later".
All of this was true. Skipping tests was arguably the right trade for a solo builder racing to validate an idea. Before AI. What changed isn't the value of tests — it's their price.
AI changed the testing math
The insight that fixed my testing problem: if AI can write features, AI can write tests. Not token tests — comprehensive suites that check edge cases a bored human would skip.
Here is my before and after, per feature. The old way: two hours writing the feature, zero tests, then four-plus hours debugging it in production. Six hours total, plus stress. The new way: 30 minutes building the feature, 15 minutes generating tests with AI, 10 minutes of debugging. 55 minutes total.
That's 6.5x faster — with tests, not despite them. The 15 minutes of AI-generated tests routinely saves me 3–6 hours of debugging, a 12–24x return on every single feature. Insurance used to be expensive; now it's almost free. Why would anyone skip it?
The exact workflow — the reusable prompt that turns any feature into a full test suite — is its own post: how I got 5,000 tests without writing them.
What good coverage looks like for a solo product
Today my platform has 5,247 tests, and Claude wrote 4,722 of them — 90%. Code coverage sits at 82%, when the industry standard for a Series A startup is 60–70%. This is not a side project; it's production software serving VCs, accelerators, and startups. And in the four months after switching to AI-generated tests, I shipped zero critical bugs to production.
But the real payoff isn't the numbers. It's what the tests let me do: refactor aggressively, ship multiple times a day, add features without fear, and sleep at night. Untested AI-generated code is exactly how the vibe coding trap snaps shut — everything works until the day it very publicly doesn't. Tests are how you build fast and build something you can stand behind when a customer calls.
Frequently asked questions
Does an MVP really need tests?
The features that touch money, data, or customer trust do. My 15-approvals bug lived in exactly that kind of feature. You don't need full coverage on day one — you need tests on the actions that would embarrass you in front of a paying customer if they broke.
How much time does testing add per feature?
With AI, about 15 minutes. Manually it was hours — which is why nobody did it. My per-feature total went from six hours (two building, four debugging in production) to 55 minutes including tests. Testing now makes me faster, not slower.
What's the ROI of writing tests?
In my numbers: 15 minutes of AI-generated tests saves 3–6 hours of debugging — a 12–24x return per feature. My pre-testing audit found 47 production bugs in six months costing 112 hours of debugging, roughly €8,960 at €80/hour, before counting lost customers.
Can AI write the tests for me?
Yes — that's what changed the economics. Claude wrote 4,722 of my 5,247 tests. AI doesn't get bored, doesn't skip "obvious" cases, and doesn't say "I'll test that later". The step-by-step method, including the reusable prompt, is in the AI-written tests post.