AI looks like magic. You type one sentence and it hands you a whole essay. You ask for a picture and thirty seconds later, there it is. And the first time this amazing thing produces a cool result, something clicks in your brain... Suddenly every boring job you hate feels like it could be gone by Friday. We know that feeling. We had it too. And we acted on it, deliberately, with the switch all the way on. This article is the report.
• We automated our entire blog with AI and zero human review. On purpose. 🤖
- The pipeline worked. The writing didn't. The AI started inventing facts and publishing them under our name.
- One article described an OpenAI programme that has never existed.
- We thought it could work, we got greedy for volume, and we made the call to run it anyway.
- The lesson: AI is a brilliant assistant and a terrible boss. Keep a human in the loop.
One of our first big ideas was simple. How hard could it be to automate our own blog? 🤔 Grab the latest AI news every day. Feed it to a robot. Let the robot write the articles. Push them straight to the website. Done.
We genuinely thought it could work. And on paper it was a double win: fresh content and SEO on one side, 📈 automation reps on the other, because building this stuff is literally what we sell. Then greed took over. More articles, more eyes, faster, and none of our time spent writing. So we made the call, fully aware of what we were doing: switch it on, no human review, let it run. We were so busy proving we could do it, we never stopped to ask if we should. 💥
Here is the machine. 🛠️ Every day it pulled stories from a stack of RSS feeds. Each story went to GPT-4o, the newest, shiniest OpenAI model at the time, with one question: would a real person actually want to read this? It scored every story out of ten.
Anything that scored seven or higher got saved to a Google Sheet. 📊 The next day a second, bigger AI agent grabbed the top five stories and wrote a full, detailed article about each one. Another step formatted it. Then an image tool called Replicate generated a matching picture. 🖼️ Finally the whole lot got pushed live to our website. No edit. No check. No human anywhere in the chain. Read that last part again. No human anywhere in the chain.
Most of it worked. The grabbing worked. The scoring worked. It was genuinely good at spotting stories people would click on. ✅ The writing is where it snapped.
The AI did not just summarise the news. It made things up. In one article it confidently told the world about an "OpenAI GenSpark accelerator", an eight-week programme with free credits, mentorship and a Demo Day. 🎤 That programme does not exist. Genspark is a completely separate company. OpenAI never ran it. The robot took two real things, smashed them together, and invented a third thing that was never real. Then it published that lie under our name. Automatically. While we slept. 😬
Here is the bit that stings. It is not that the AI was dumb. It is what we chose to automate. A blog is a trust machine. 🤝 Someone gives you ten minutes of their life to read what you wrote, and in return they expect you not to lie to them. That is the whole deal.
We took that deal and handed it to a robot with zero oversight, because we wanted volume. That was greed, and it was a deliberate choice, and those two things can be true at the same time. The only thing we did right was the target: we ran it on our own blog, where the cost of failure was our embarrassment and nothing else. If this pipeline had been pointed at a client's website, the cost would have been their reputation. 🔥
There is a well-known trap where people who just learned something feel like experts, right before they find out how much they do not know. 📉 Every business getting excited about AI right now is somewhere on that curve. We already rode it. All the way up, and all the way back down. This blog is the wreckage.
Here is what we know now that we did not then: AI is a brilliant assistant and a terrible boss. The second you hand it full control of anything that touches trust, your writing, your numbers, your customers, it will eventually embarrass you. Not because it is evil. Because it does not know what it does not know, and it never stops to ask.
The fix is not "use less AI." The fix is a human in the loop. 🧯 Someone who checks. Someone who is accountable when it is wrong. That is the difference between automation that builds your business and automation that quietly burns it down. That is the whole job we do now: we build the automation, and we build the checks. We already paid for this lesson so you do not have to.
Want to see what unsupervised AI actually produces? We left the articles up on purpose. 👇 They are messy, they oversell random tools, and at least one of them is lying to you. Go and spot the hallucinations.
➡️ AI and no-code: South African SMEs' blueprint for growth
➡️ AI empowers lean teams to outsell corporate giants
➡️ Democratised AI for SA SMEs: DIY audiobooks and GPT prototypes
➡️ Drag, drop, deploy: no-code AI transforms SA SMEs
➡️ Funnels, emails, ads and social workflows
➡️ Hyperautomation for South African SMEs
➡️ Low-cost AI automations: spreadsheets, drips and GPT assistants
➡️ SA SMEs automate everything with Make.com's no-code AI
➡️ SA SMEs embrace free AI for faster sales, support and insights
➡️ SA SMEs embrace no-code multimodal AI: chatbots to video
➡️ SA SMEs harness AI for forecasting, support and design
➡️ Templates and AI tools cut SA SMEs' project timelines
➡️ Zapier and no-code AI let SMEs punch above their weight
Brainstormed with AI. Written by a REAL human.
