An AI agent that writes and posts LinkedIn for a founder should do three things and refuse a fourth. It should take the next topic from a calendar the founder wrote, write the post to rules the founder set, and keep a record of every run, the failed ones included. What it should refuse is to click Post on the founder's personal profile without the founder reading the draft, because LinkedIn's terms do not permit automation at all and a founder's profile is the one account that cannot be replaced if it is restricted. Ours runs on our own company page, where we chose that risk for ourselves: 30 text posts between September 26 and October 31, and 3 videos since October 24. This is what it writes to, what came back, and the one step we would hand back to a founder.
What GPT-6 Sol's search returns for this question
On the morning of November 1 we put the question through GPT-6 Sol's web search and kept the list: 31 results, none of them ours and none a Reddit thread. Twelve were product pages: SignalPost, Never Boring AI ("your AI agent that writes LinkedIn posts"), PublishFlow ("AI LinkedIn Ghostwriter for Founders"), Growli, Postiv's Bob, Allen, ItGrows, Sacha, Renly, Imagine AI, Typoro and Sevenfold. One was a TechRadar piece whose title quotes LinkedIn's own position: "It's ok to use AI to help you write, but your posts and comments need to represent your voice and your perspectives", with the words "crack down on AI slop" after it. We read the titles only. Twelve products offer to write and post for a founder; the platform, in the same result list, says the posts have to be the founder's own. The rest of this post is how we hold both at once.
The routine, as our two LinkedIn agents run it
| Text posting agent | Video posting agent | |
|---|---|---|
| What it takes | The first pending row of a 30-row calendar: a topic and an angle | The oldest video of ours that is public on YouTube and not yet on LinkedIn |
| What it writes | A hook as the first line (about 200 characters show before "see more"), short conversational paragraphs, 600 to 1,300 characters, no links, no dashes, two to four hashtags on the last line, no call to action | A post text for the video, read by our content review before anything posts, with one revision allowed |
| How it posts | Types the post into the company page's composer in a dedicated Chrome kept signed in, attaches a 1080 by 1080 card drawn from the first line, clicks Post, then adds the backlink as a comment | Uploads the native video to the company page through the same kind of Chrome, Monday to Friday after 12:30 US Eastern |
| Who reads it first | Nobody, in our own run: the calendar, the rules and a daily cap are the controls | The content review, which has sent texts back |
| What it records | The calendar row flipped to posted with the date | A registry entry per video, with impressions, members reached and video views read from the page's analytics at two and seven days |
| What it does when a page element is missing | Stops, reports the error with a screenshot and leaves the row pending for the next run | Stops the same way, and since October 24 it may repair its own script and commit the repair for a person to see |
The second column exists because the first has a gap. The text agent's calendar records the date each row was posted and nothing about what the post did; LinkedIn offers no posting API the agent could use, as its notes put it, so the weekly report reads our X posts through the API and reads nothing from the company page's posts. The video agent was built to read its own analytics page, and since October 31 it does.
An AI agent that writes and posts LinkedIn for a founder: what a month of ours produced
| Count | |
|---|---|
| Text posts | 30, from September 26 to October 31; 29 in October |
| October days with no post | 4 (October 1, 5, 7 and 17); 2 days carried two posts (October 6 and 18) |
| Posts in the week to October 30 | 6 |
| What the text posts drew | Not in our records |
| Videos posted | 3: October 24, 26 and 29 |
| The first video at seven days | 103 impressions, 64 members reached, 35 video views, 4 clicks, 1 follow |
| The second video at five days | 9 impressions, 3 members reached, 6 video views |
| Replies to our 32 LinkedIn messages to founders since September 21 | 0, as of October 30 |
Three readings. First, the output is the easy part: once a calendar of 30 rows exists the posts go out, and the four empty October days show in the calendar as gaps between dated rows, with a blank where the post's link would be because the script does not capture it. Second, the only reach figure we have for the page is the video agent's, and it says what a new company page gets: the first video reached 64 members in a week and the second 3 in five days. Third, the last row is a different channel, direct messages a person sent, and it is in the table because a founder asking this question usually wants customers, and a page posting every day for a month put no number in that row.
The step a founder should keep
The rules a post is written to are a founder's to set, and ours are the ones we would hand over first: the hook on its own line, no links in the body, the 600 to 1,300 character band, two to four hashtags. A founder who cannot write that list on one page has no routine for an agent to run yet, and the agent will post the wrong thing on time.
The click is the step to keep on a personal profile. On our own company page the agent clicks Post, at a human pace with a daily cap, and we accept the chance that LinkedIn objects. For a founder's own profile the same agent stops one step earlier: the draft is written to the founder's rules and the founder reads it and posts it, or does not. That is also the only arrangement that meets the platform's own line, quoted in the TechRadar title above, that the posts have to represent the member's voice. A ghostwriter who publishes under your name without you reading the piece is not a service we would buy for ourselves, and we do not sell it for a profile.
What the agent should never do is pick the topic. Ours reads a calendar top to bottom and has never chosen a row; when the 30th row was posted on October 31 the next run stopped and said the calendar needed rows. The founder's judgment lives in the calendar, and the LinkedIn posting agent's page shows the rows as they flip.
How we checked
The text counts are from the agent's topic calendar, 30 rows each with the date it was posted, read on November 1; the week's count of 6 is from our weekly performance record of October 30, which also holds the LinkedIn message figures (164 connection requests, 60 accepted, 32 messaged, 0 replied since September 21). The video figures are from the video agent's registry, three entries with the counts its analytics reader stored on October 31. The rules are from each agent's workflow page. The search results are from our GPT visibility round of November 1, one search for this question with the list kept; we did not open any product page in it, so no vendor's price or claim is quoted.
Disclosure
ClawWorld runs AI growth agents for early-stage startups and small businesses: they carry out the growth routine a team already planned (posting, replying in communities, outreach, launches and measurement) on a schedule, with every run on a public log, and AEO (answer engine optimization) is one of the services. The two LinkedIn agents in this post are among them, free to copy from their pages or priced per agent to have us run them, so we have an interest in this question. Every number above is from our own records, and the row we do not have is marked as such.