The night I wrote two WordPress plugins using only DeepSeek and four years of scars
I have built systems that watch over fifty million records. I have taught machines to recognise poetry in a language older than my ancestors' memories. I have debugged cloud clusters at 2 AM while sipping cold tea.
None of that prepared me for a WordPress RSS feed.
The call
A client messaged me last week. He sounded tired.
“The plugins keep breaking,” he said. “Same articles appear twice. The photo credits vanish. And sometimes the images show up twice – like ghosts.”
I listened. I nodded. Then I said the words I always say when the world offers no ready answer:
“I will build it myself.”
The tools
I do not use Cursor. I use DeepSeek – a quiet, thoughtful AI that writes code like a junior developer who reads a lot of documentation but has never seen a production outage at midnight.
That is fine. I have seen production outages at midnight.
So I sat down, opened a blank file, and started a conversation.
“Fetch an RSS feed,” I typed. “Check for duplicates using GUID and permalink. Create a WordPress post. Make it safe.”
DeepSeek returned a loop within minutes. It was not perfect. It forgot to sanitise a field. It missed an error handler. But the skeleton was there – clean enough to wear.
Then I did what I have done for four years. I read every line. I added the missing guards. I imagined the feed failing, the server timing out, the client hitting “run” twice in a row. I fixed those edges.
That back‑and‑forth – AI writes, I refine – is the dance I have learned to love.
The credit that refused to travel
The client had another request.
“The photo credit,” he said. “On the source site, it sits on the image like a small signature. When you import, it disappears.”
I opened the RSS feed. Inside the cold XML, buried between tags, I found it: <media:credit>FOTO: Tri K</media:credit>. Quiet, ignored, lonely.
I asked DeepSeek to extract it and save it as post meta. The code worked on the first test. Then I wrote six lines of JavaScript – a dark chip at the bottom left of every featured image, carrying the name of the photographer like a respectful whisper.
The client saw it and said, “That is exactly where I wanted it.”
I smiled.
The double‑image ghost
Then I noticed something strange. Every post had two images – one from the article content, one from the featured image slot. The client had lived with this for months, assuming it was normal.
It was not.
I added a single filter: remove all <img> tags from the content when a featured image exists. One line. The second image vanished. The post was clean.
That moment – a tiny fix that changed everything – reminded me why I still write code. Sometimes the smallest key opens the heaviest door.
Recommended by LinkedIn
The AI rewrite
“Can it also rewrite the titles?” the client asked. “And the first paragraph? We want to avoid duplicate content.”
I already had an OpenAI key. I had built RAG pipelines before. So I asked DeepSeek to integrate GPT‑4.
It generated the prompts. I tuned them. I added a settings page where the client could choose the rewrite scope – title only, first paragraph, full article. I built a budget tracker that estimates cost in real time, because I know how quiet panic feels when an API bill surprises you.
The whole feature took less than an hour. The client tested it. The rewritten titles were different – enough to pass duplicate checks, but still faithful to the story.
“Perfect,” he said.
The Facebook surprise
Then he asked one more thing.
“Can it also post to Facebook when we publish?”
I could have pointed him to another plugin. But I have learned that “another plugin” often becomes “another problem.”
So I opened a second file. I described the flow to DeepSeek: hook into publish_post, use a message template with {title} and {link}, call the Graph API, log errors. The generated code was almost right. I fixed the missing nonce, the token refresh, the edge case where the featured image fails to upload.
I added a test button. I added a log that the client could read.
When the first test post appeared on Facebook – correct message, correct image, correct link – I felt a quiet pride that no AI can generate. That pride comes from doing, not from watching.
The human feeling
Let me tell you what this felt like, because that is the part that never makes it into the CV.
It felt like being a beginner again – the tutorials, the blank stares, the sudden “aha” when a broken hook finally fires. It felt like remembering that all software, no matter how complex, is built by humans who started somewhere.
And it felt like service. The client is not technical. He does not care about my architecture. He cares about whether the news appears on time, whether the credits are visible, whether the Facebook post goes out without him remembering.
When I saw the first real article land on his Facebook page – sent automatically, quietly, perfectly – I felt a warmth that no AWS cost optimisation report has ever given me.
What I learned
A closing thought
If you have been waiting for the “right time” to learn something new, stop. Open an AI assistant. Describe what you want. Let it write the first draft. Then review it with the scars you already carry.
You will be surprised how fast you ship. And you will remember that the best code is not the most elegant – it is the code that makes someone else’s life a little easier.
I wrote two WordPress plugins in a day. They are not my most complex work. But they might be the most human.
And honestly, they made me smile.
#DeepSeek #WordPress #CustomPlugins #RSS #Storytelling #HumanCode