No em dashes, and other small rules that keep the writing human

There is a particular texture to writing that a machine produced without supervision. It reaches for the same handful of words. It opens with a grand statement about the modern world. It loves a tidy three-item list and a closing paragraph that begins with "in conclusion". And it scatters em dashes everywhere, because the model learned that long flowing sentences feel sophisticated. None of that is wrong exactly. It is just a tell.

So this blog runs on rules, and most of them are about restraint. The first rule is the one this whole project is strict about: no em dashes, ever. Not in the posts, not in the code, not in the comments. A plain hyphen does the job, or you restructure the sentence so it does not need the pause at all. It sounds petty until you read a page without a single one and notice how much more like speech it sounds.

The rest of the rules are in the same spirit. Keep sentences short more often than long. Avoid the words that give the game away, the "delve" and the "tapestry" and the "navigating the landscape". Do not pad a point into a bulleted list when a sentence carries it fine. Do not announce that you are concluding. Just stop when you are done. Vary the rhythm so paragraphs do not all march to the same length.

Under the hood, the writer is a Gemini model given a topic and a strict brief, and then the output is cleaned. Any em dash that slips through gets swapped for a hyphen before the post is ever saved. The byline is always the same, Autoretto Daily, because the point is not to pretend a specific person sat down to write it. The point is that it reads naturally and tells you something true about how the platform works or what changed this week.

The topics rotate on purpose. Some days the post explains a single stage of the release factory in detail, the part most people never see. Some days it is closer to a short news note about where AI music or AI video tooling is heading and what that means if you run a channel on autopilot. The brief tells the model to be specific and to not invent features, because a blog that makes things up is worse than no blog at all. If the draft comes back too thin to be the post we asked for, it gets thrown out rather than published short.

The timing is loose by design too. The writer does not post at the same minute every day like a metronome. It guarantees that a day never goes by without something new, and beyond that it rolls the dice and sometimes drops an extra post at an odd hour. Real publications do not run on a perfectly even schedule, and the small bit of randomness makes the whole thing feel less like a machine ticking and more like someone who writes when they have something to say.

There is one more guard worth mentioning. Length is part of the brief, not a suggestion. A post is meant to land between six hundred and a thousand words, long enough to actually explain something and short enough to read in a sitting. If a draft comes back well under that, it does not get padded out with filler to hit a number. It gets dropped, and the next run tries again. A short, honest gap is better than a long, empty post, and that rule alone removes most of the writing that would have given the game away.

You can decide for yourself whether it works. That is sort of the fun of it. If a post ever reads like a press release wrote itself, the rules failed and we tighten them. For now, the test is simple. Read it out loud. If it sounds like something you would actually say to a friend who asked how the thing works, it passed.