I truly dislike LLM-generated writing. The quality of the models’ output has vastly improved over the years, but the result stays the same. Writing is deeply human. Putting our thoughts on paper, with all the struggle and the rework, is how we learn. It’s how we get better as individuals and as a species.
So, every autogenerated newsletter and LinkedIn reply feels like an affront to humanity. And it’s getting worse by the day.
We marvel at Moltbook, but most of X is bots talking to bots. Half of my e-mail is auto-generated cold outreach. My entire spam folder is written by AI. TikTok, YouTube, and Instagram run on autopilot with machine-spawned drivel flooding the channels. In the very near future, most of our texts, WhatsApp messages, and phone calls will come from spammy machines owned by humans who are too lazy to do the work.
So, please, for the sake of humanity: write your own articles, replies and thinkpieces. I don’t care whether ChatGPT has a better writing style than me. I write for my own growth and hopefully for a little bit of yours, too.
Lately, however, I’ve been working on a kind of automated writing that is an absolute delight: technical tutorials and reports.
Software engineers have a prolific history of writing and reading technical articles. Whenever a dev needs to understand something, they type “How do I…” in Google to discover a myriad of articles, videos and StackOverflow posts. Outside of software development, we see similar knowledge sharing. “How to integrate HubSpot with Gmail?” gives you a treasure trove of insights.
None of these articles are page-turners. None of these videos are Oscar-worthy. But it is a valuable way for humans to share knowledge in their own technical field.
Writing tutorials takes time and effort. I applaud every engineer who has written a free 8-page primer on some specific Java 5 quirk for an audience of 10. It’s high effort/high impact, but with a very tiny target group.
There is another issue with documenting after the fact. We tend to forget struggles and details that mattered. In my article about vibecoding in Malborge, I recounted my attempts from memory. I didn’t meticulously keep track of all the changes, tweaks and roadblocks encountered along the way. It’s exactly these struggles that make it valuable for the audience. I told them what worked, not how I learned why.
You know who is very good at keeping such a log? Your trusty LLM, of course.
Modern LLM tools build context. They don’t forget the conversation after each reply. That doesn’t come naturally to large language models, by the way. That kind of memory is a feature in the tools built on top. ChatGPT uses conversation history as specific memory and stores some information about you in long-term memory. Coding agents build a plan and write it to disk to remember what they have tried and why. OpenClaw tweaks and stores its personality in a SOUL.MD file.
This context is a superpower because it makes ChatGPT and Claude Code feel like smart agents rather than dumb API calls. But it also turns them into scientists with OCD. They log and remember what we tried and why we came to a conclusion. And given the fact that they are proficient writers these days, one plus one clearly makes two.
“Hey Claude, write what we learned in a technical tutorial and publish it on GitHub”
My experiments with generating sprites led to this dry article.
Fooling around with different CLI agents resulted in this funny, but insightful repo.
But when diving real deep, the agents shine. My benchmark of CPU-based small LLMs has detailed reports that give peers the tools to verify and spark conversation.
All of these articles have been written by AI. They capture my learnings and share it with the world.
I feel more of us should do that. These tiny bricks add to the human knowledge base in a way that can be consumed by people and machines alike.
Humans should write to explore and shape their thoughts.
Machines should write to gather and summarise the artefacts of human exploration.
Both should share those findings freely.
So, next time you have a long back-and-forth with ChatGPT about a topic in your field, don’t close the conversation and lose the precious context.
Ask it to write a technical article and share it with the world.
