Social Comments

Three times a day, finds relevant videos from other creators and leaves one genuinely useful comment on each platform.

Only comment when you are certain. Being wrong in public costs more than being quiet.

Audience growth for a channel, done the slow honest way: find videos actually worth engaging with, watch what they are about, and leave one comment that adds something.

How it is built

The hard constraint is reading the content first. The AI agent pulls the transcript or description and must find something specific to respond to. If it cannot, it does not comment. A generic compliment is worse than silence, it reads as a bot, because it is one.

Tone is tuned per platform. Comments are short, colloquial, and never claim an experience that is not mine.

What it got wrong

It invented experience. An early version wrote comments implying first-hand knowledge of watches I have never handled. Plausible, well-written, and completely fabricated. The rule now is explicit: no unverified facts, no invented experiences, and when in doubt, do not post.

It typed into the wrong element. On one platform a stray keystroke before focus lands is a shortcut, not text, meaning a failed comment could silently like a video instead. Every run now confirms focus before typing.

The lesson

The reputational risk of an autonomous AI agent is not that it breaks something. It is that it says something confident and wrong under your name, politely, at scale.