I Built My Mum a Bullshit Detector and Discovered My Real Bug

My parents get an uncountable number of WhatsApp forwards a day. Articles with headlines designed to make a vein in your forehead twitch, linking to sites I have never heard of. Half of them claim vaccines cause autism. The other half insist lemon water cures cancer. As the family's designated tech person, I became the unofficial fact-checker by default, which is a job with no salary and a hostile customer base.

"Mum, don't trust that site, it's super biased."

"Dad, that's The Onion. It's satire."

"Please stop sharing articles from whatever-dot-freedom-eagle-news."

After enough of those conversations I decided to build an AI one instead, mostly so I could be ignored by a machine instead of in person.

The build that should have taken an afternoon

The idea was not complicated. Paste in a URL, get back an analysis of the source's bias and credibility. I used v0 to build the frontend and Make to handle the AI workflow. The core AI bit is not sophisticated. It is an LLM with a system prompt that scrapes a page and looks for bias indicators, which is roughly the technical complexity of a smoke alarm.

The speed at which I could cobble it together did surprise me. I could have built the entire thing in a few hours.

Could have.

The reality is it took several days, because every time I got one piece working, something else broke, or my greedy mind asked "but what if it also did this one other thing?". As I write this I am still telling myself I should hold off sharing until I make ten more changes. The bottleneck was never the AI. It was the LLM's inability to handle backend infrastructure colliding with my inability to leave anything alone. Every superpower has its own shadow, and mine apparently has the impulse control of a Labrador near a sandwich.

What AI actually did well

The scraper uses ScrapingBee's AI-powered extraction, which takes natural language descriptions of what to pull. Instead of writing CSS selectors or parsing logic, you tell it "return a list of products and their prices" and it works out the rest from any site layout. No custom scrapers for every site structure.

Content processing turned trivial. The AI took messy copy-pasted JavaScript and returned clean analyzable text without me writing a parser. Database integration, the tedious work I always procrastinate on, it structured and pushed efficiently. v0 turned rough ideas into working React components faster than I could type them.

At no point was the AI replacing my judgment. It did not decide what mattered to a user or what the product was for. It removed the friction between wanting a thing to exist and the thing existing, which is a smaller claim than the marketing makes and a much more useful one.

The trap nobody warns you about

The "AI will revolutionize development" discourse skips the quiet part. When building something took weeks, you had natural stopping points. You shipped because you were exhausted. Now every extra feature feels like it is one prompt away. "Oh, I could add sentiment analysis." "What about tracking source reliability over time." "Maybe it needs a Chrome extension."

The tool could have solved my original problem at version 0.1. I kept going, not because the features were necessary, but because they were possible.

That creates a new kind of debt. Spin up features quickly and you end up maintaining a Frankenstein of half-baked additions. AI helps you build faster. It does not help you decide what not to build, and if anything it makes "just one more thing" more dangerous because the cost feels close to zero. Old technical debt came from shortcuts. AI-era technical debt might come from building too much, too fast, with no architectural thinking holding it together.

What actually mattered

Despite all the bells I bolted on, the core value stayed simple. My mum now has somewhere to check articles before sharing them, and in some small way I am maybe reducing the spread of nonsense through WhatsApp forwards.

The technology is not revolutionary. The AI is not clever. The collapse in distance between idea and execution is the genuinely new thing, and it is more disruptive than people admit precisely because it is so unglamorous.

We are not in the age of AI replacing developers. We are in the age of AI making developers unreasonably productive, which is the more interesting problem because productivity with no off-switch is just a faster way to overcomplicate things.

So before your next build, write down the one problem it has to solve and the version where it is allowed to be finished. The AI will not set that boundary for you. That part is still yours, and now I just have to get my mum to use the thing before she forwards me a cure for cancer.