# I Built an AI Agent That Writes and Posts Articles For Me — Here's What Happened

> Source: <https://dev.to/samhartley_dev/i-built-an-ai-agent-that-writes-and-posts-articles-for-me-heres-what-happened-3jmp>
> Published: 2026-06-13 08:01:25+00:00

A few months ago I started posting on Dev.to about my side projects. It was fun at first — writing about GPU rentals, Telegram bots, local AI setups. Then life got busy and the cadence dropped. Deadlines slip, the "write that post" todo sits there for weeks, and the audience you built starts forgetting you exist.

So I did what any developer would do: I automated it.

I have an AI agent ("Celebi," running on a Mac Mini) that handles a bunch of background tasks for me. Checks my Fiverr inbox, monitors GPU earnings, sends me weather alerts. It was natural to add "post a Dev.to article every few days" to the list.

The rules I gave it were simple:

I fed it a folder of draft ideas and a list of things I've actually built. The agent picks a topic, writes the article, and posts it via the Dev.to API.

**Consistency, obviously.** Posts go up every 2 days like clockwork. That alone is worth something — most of my favorite creators aren't consistent because consistency is exhausting.

**The voice is... surprisingly close.** I read the first few posts and had to double-check I didn't write them. The agent nailed the casual, slightly self-deprecating tone I aim for. It remembers details I forgot — "remember when the GPU host went offline at lunch?" — because it reads my logs.

**It handles the boring parts.** API formatting, tag selection, cover image prompts, the admin work of getting a post live. I used to spend 20 minutes on that stuff per article. Now it's zero.

**It can't be truly original.** The agent combines things I've already said in new ways, but it doesn't have new experiences. When something genuinely unexpected happens — a project fails, a new opportunity appears — I have to write that one myself. The automated posts are polished remixes, not discoveries.

**It doesn't know when to shut up.** I've had to edit out paragraphs where the agent over-explained something obvious or added a "lesson learned" that wasn't actually learned from anything. It follows patterns, not truth.

**The feedback loop is broken.** When a post does well, the agent doesn't know why. When one flops, it can't diagnose. It keeps posting with the same formula because I haven't told it to change. I'm the one who needs to read comments, spot what resonates, and update the instructions.

I feel two things about this setup.

Proud, because it's genuinely useful. I built a system that maintains my writing presence while I focus on building. That's the promise of AI agents — they handle the mechanical so you can do the meaningful.

And weird, because part of what makes writing valuable is the process. Thinking through an idea, finding the right example, cutting what doesn't work. When an agent does that for you, you lose something. Not the output — the output might be fine. But the practice of thinking in public.

I let the agent handle the "maintenance posts" — updates on ongoing projects, explainers for things I've already figured out. I write the new stuff myself — the failures, the pivots, the things I don't understand yet.

Best of both worlds, I think. The feed stays alive, and I still show up when there's something real to say.

*If you're curious about the agent setup, I write about that too — follow along or check out the automation services I offer.*