Draft and publish LinkedIn posts from voice notes
#ai. Task: #ai #agents #grok bot Grok Bot just dropped. I downloaded it on a Mac the same day and recorded myself setting it up. This is the getting started version of that. Not a feature list. What I actually clicked, what worked, and which bots I would create if I were you. What it actually is It is a team of bots on a computer that is not yours. They do the stuff you would hand a teammate. Inbox, LinkedIn, GitHub, the lot. The marketing page shows sample teammates like talent scout, account manager, inbox manager. That is the idea. You do not start with twelve of those. You start with one bot and a name. On first run you can pick a colour and a role. Coding and repos. Research and writing. Inbox and calendar. Or a bit of everything. If you do not know what you want, it gives you options and you click. That onboarding is the best bit. Name the bot. Answer a couple of questions. It guides you. I already had a first bot, deleted it, and started again on camera. So your first screen may look a little nicer than mine. Bot 1: a coding bot I created a new bot and picked coding and repos, just to see how far it would go. It asked where the code lives. GitHub. It already had a GitHub connector installed from when I poked at plugins before recording. Sign-in hit a snag. It offered another GitHub connector that wanted a personal access token. I said I would do that later. Cloud agents still work if GitHub is linked in Cursor, and mine is. Then it asked what to jump on first. Ship features, review PRs, explain codebases, or a mix. I picked a mix. A few core repos, not everything. I could not remember the exact repo name. I typed "playwright movies". It found debs-obrien/playwright-movies-app from my GitHub. From there I asked some simple stuff. Open issues. Stars. Bring up issue 29, the timeout one. Nothing magic yet. You could do that in any chat with a GitHub MCP. While that ran I created a second bot. You can have more than one. That is the whole point. Bot 2: a LinkedIn bot I named it LinkedIn bot. That was enough. It already knew I meant LinkedIn. It asked what I wanted: draft posts and comments, polish the profile, job search, outreach. I picked draft posts and comments. Then how to write: warm and conversational, match my existing posts, be me. I picked match my existing posts. Then the important bit. It opens a computer that is not yours. You sign in there. It never sees your password. You get a banner: sign in, do 2FA if asked, then "I'm done, continue". It feels a bit like someone remote-controlling a tiny desktop. Once you are in, you hand it back. I flicked back to the coding bot while LinkedIn was signing in. The coding bot had checked issue 29, found the tests already used waitForURL and no hard waits, and asked if it should close it. I said close it when GitHub is connected. Then I signed GitHub in on that computer. It was already signed in as debs-obrien. It closed the issue with a note. I clicked through to GitHub to see it for myself because I did not believe it actually did it but it did. Back on LinkedIn, it had pulled recent posts and locked in how I write. Conversational build-log. Concrete numbers. I dumped a messy voice note: I am recording a video of setting this up, I now have a LinkedIn bot, this is real. It drafted in my voice. I said post it. It took a minute. Long enough that I was sure it would fail. Then: it is live. Opening with "I am literally writing this post from Grokbot right now." It opened the post on its computer. Impressions already ticking. I tried to add a screenshot after the fact. LinkedIn will not let you attach an image to a post that is already live. Edit is text only. The bot offered delete and repost, or leave it. I left it, then asked it to put the screenshot in the first comment. That worked. So: name the bot, sign in once on its computer, dump a thought, review the draft, post it. Plugins There is a Plugins item in the sidebar. Gmail, Google Calendar, Google Drive, Notion, Slack, Playwright, GitHub, X, a pile of others. I had already added a couple before I hit record, which is why GitHub was half-connected. Click along. If you do not know what you need, this list is the menu. Bot 3: email I created an email bot. I am not showing you my inbox in a YouTube video. Inbox triage, drafts, digest, or all of it. I picked all of it. Gmail was already there. Connect, pick the account, allow access. It pulled unread vs read in seconds.; tools: Grok Bot. Human role: The original post does not say how a person stayed involved.; limitations: LinkedIn's native API or UI restrictions prevent adding media to published text-only posts, necessitating manual or bot-driven comment additions. The bot operates on a remote computer, requiring the user to handle sign-in and 2FA securely without sharing passwords. Publishing latency can be noticeable, and the system relies on the bot's ability to accurately mimic the user's writing style from existing posts..
Tools
Published on Agent Work Atlas:
Overview
#ai
Workflow
#ai #agents #grok bot Grok Bot just dropped. I downloaded it on a Mac the same day and recorded myself setting it up. This is the getting started version of that. Not a feature list. What I actually clicked, what worked, and which bots I would create if I were you. What it actually is It is a team of bots on a computer that is not yours. They do the stuff you would hand a teammate. Inbox, LinkedIn, GitHub, the lot. The marketing page shows sample teammates like talent scout, account manager, inbox manager. That is the idea. You do not start with twelve of those. You start with one bot and a name. On first run you can pick a colour and a role. Coding and repos. Research and writing. Inbox and calendar. Or a bit of everything. If you do not know what you want, it gives you options and you click. That onboarding is the best bit. Name the bot. Answer a couple of questions. It guides you. I already had a first bot, deleted it, and started again on camera. So your first screen may look a little nicer than mine. Bot 1: a coding bot I created a new bot and picked coding and repos, just to see how far it would go. It asked where the code lives. GitHub. It already had a GitHub connector installed from when I poked at plugins before recording. Sign-in hit a snag. It offered another GitHub connector that wanted a personal access token. I said I would do that later. Cloud agents still work if GitHub is linked in Cursor, and mine is. Then it asked what to jump on first. Ship features, review PRs, explain codebases, or a mix. I picked a mix. A few core repos, not everything. I could not remember the exact repo name. I typed "playwright movies". It found debs-obrien/playwright-movies-app from my GitHub. From there I asked some simple stuff. Open issues. Stars. Bring up issue 29, the timeout one. Nothing magic yet. You could do that in any chat with a GitHub MCP. While that ran I created a second bot. You can have more than one. That is the whole point. Bot 2: a LinkedIn bot I named it LinkedIn bot. That was enough. It already knew I meant LinkedIn. It asked what I wanted: draft posts and comments, polish the profile, job search, outreach. I picked draft posts and comments. Then how to write: warm and conversational, match my existing posts, be me. I picked match my existing posts. Then the important bit. It opens a computer that is not yours. You sign in there. It never sees your password. You get a banner: sign in, do 2FA if asked, then "I'm done, continue". It feels a bit like someone remote-controlling a tiny desktop. Once you are in, you hand it back. I flicked back to the coding bot while LinkedIn was signing in. The coding bot had checked issue 29, found the tests already used waitForURL and no hard waits, and asked if it should close it. I said close it when GitHub is connected. Then I signed GitHub in on that computer. It was already signed in as debs-obrien. It closed the issue with a note. I clicked through to GitHub to see it for myself because I did not believe it actually did it but it did. Back on LinkedIn, it had pulled recent posts and locked in how I write. Conversational build-log. Concrete numbers. I dumped a messy voice note: I am recording a video of setting this up, I now have a LinkedIn bot, this is real. It drafted in my voice. I said post it. It took a minute. Long enough that I was sure it would fail. Then: it is live. Opening with "I am literally writing this post from Grokbot right now." It opened the post on its computer. Impressions already ticking. I tried to add a screenshot after the fact. LinkedIn will not let you attach an image to a post that is already live. Edit is text only. The bot offered delete and repost, or leave it. I left it, then asked it to put the screenshot in the first comment. That worked. So: name the bot, sign in once on its computer, dump a thought, review the draft, post it. Plugins There is a Plugins item in the sidebar. Gmail, Google Calendar, Google Drive, Notion, Slack, Playwright, GitHub, X, a pile of others. I had already added a couple before I hit record, which is why GitHub was half-connected. Click along. If you do not know what you need, this list is the menu. Bot 3: email I created an email bot. I am not showing you my inbox in a YouTube video. Inbox triage, drafts, digest, or all of it. I picked all of it. Gmail was already there. Connect, pick the account, allow access. It pulled unread vs read in seconds.
Reported result
The bot successfully drafted and published a LinkedIn post from a voice note, with impressions beginning to accumulate. However, the author encountered a limitation where LinkedIn does not allow attaching images to already-published posts, requiring a workaround to add the screenshot as a comment. The publishing process took approximately one minute, causing initial concern about failure.
Human role
The original post does not say how a person stayed involved.
Limitations
LinkedIn's native API or UI restrictions prevent adding media to published text-only posts, necessitating manual or bot-driven comment additions. The bot operates on a remote computer, requiring the user to handle sign-in and 2FA securely without sharing passwords. Publishing latency can be noticeable, and the system relies on the bot's ability to accurately mimic the user's writing style from existing posts.
Try it
Act as my LinkedIn content assistant using Grok Bot. First, create a bot named 'LinkedIn Bot' and configure it for 'draft posts and comments' with the style 'match my existing posts'. I will provide a raw voice note or text dump about [replaceable topic, e.g., a recent project update]. Draft a LinkedIn post in my voice. Show me the draft for approval before publishing. If I approve, publish the post. If I want to add an image [replaceable image file path], note that LinkedIn may not allow editing published posts to add media; instead, post the image as the first comment immediately after publishing. Verify the post is live and the comment is attached. Do not publish without my explicit confirmation.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.