Back to the letter

Research report · 14 Nov 2026 · Status: testing from today

I automated my LinkedIn for free by merging three open-source projects

I compared the open-source and paid ways to run a LinkedIn account with AI, kept the best parts of two MIT skill bundles and Agent Reach, added a posting robot on LinkedIn's free official API, and start a one-week test today.

In short

Why I looked at this

My feed is full of AI slop, and I don't enjoy scrolling it. But I still want to reach founders and builders. So I wanted a system that does the slow parts for me and still sounds like me, not like the posts I'm tired of reading.

My usual method: find every free and paid tool in a space, read how each one works, then combine the good parts into my own setup. That gets me most of what companies charge for, at close to zero cost.

The tools, compared

ToolWhat it doesFree or paidLicenceWhat I tookWhat it lacks
Jake Schincariol's linkedin-agent-skill11 short Claude skills, including a humanizerFreeMITThe structure and the runnable humanizer scriptsNo posting or reading. You paste everything.
Serge Bulaev's linkedin-skills12 skills with deeper research: hooks, feed notes, interviewer, thread monitorFree, optional paid add-onsMITThe knowledge: hooks, Humanizer V3 rules, the interviewerReading and posting depend on Apify and Publora
PubloraPublishing API for LinkedInFreemium (15 posts a month free, per Serge's README)ProprietaryNothingA third party gets my account access. The official API does it free.
Apify LinkedIn actorsScrapers for posts, comments, engagersPay per result ($1 to $5 per 1,000)ProprietaryNothing. Agent Reach replaced it.Ongoing cost
Agent ReachOpen-source CLI that lets an agent read 16 platformsFreeMITAll LinkedIn readingReads as my logged-in account, which LinkedIn forbids
Share on LinkedIn APIOfficial permission to post as yourselfFreeLinkedIn API termsAll posting, through my own scriptNeeds an app tied to a company Page
Community Management APIOfficial likes and commentsGatedLinkedIn API termsNothingOnly for registered organisations

I dropped the idea of a browser bot that clicks Like for me. It is the clearest breach of LinkedIn's rules. For price context: paid LinkedIn tools start at roughly $20 a month (AuthoredUp) and $39 a month (Taplio), going by their published pricing.

What I built

  1. Interview. /li-interview collects my real stories and numbers. A missing number stays a {{my number}} slot, never a guess.
  2. Plan. /li-plan lays out 7 posts across 7 daily pillars I rebuilt for a founder voice, like Build Log and Startup Teardown.
  3. Write. /li-post gives three hooks from a 40-hook library, one draft, and an image prompt when useful.
  4. Humanize. /li-human runs two local scripts. A deliberately bad post scored 7.7 (flagged), a clean one 86.7 (pass).
  5. Approve and publish. Nothing goes out until I type "approve". A Windows task checks the queue every 15 minutes and posts through the official API. My own part.
  6. Engage. /li-monitor and /li-engagers read through Agent Reach. Claude drafts comments. I click Like and paste them myself.
  7. Audit. /li-audit ranks posts by engagement rate, not impressions, and feeds next week's plan.

My own additions: the posting robot, the founder pillars, image prompts, auto-reshares (max one a day), the read cap and a 25-test suite. What broke: the humanizer crashed on the rupee sign because my Windows console uses cp1252, and python3 on my PC is a Microsoft Store stub. Both are fixed.

What it replaces and what it costs

It replaces a scheduler, an AI writer, a post analytics tool and a scraper, with no extra tool fees. It runs on Claude Code on the Claude Pro plan, which I already pay for. So it is free to add, not free to run.

It does not replace tools that auto-like and auto-connect. I chose not to copy that part. So it is not "touch nothing": I still spend about 10 minutes a day on likes and comments, plus approving each post.

The test, starting today

One week, from 14 Nov 2026. Before it started, I published a post saying this account is running an AI automation experiment. Every post is still read and approved by me.

What I will measure, per post and for the week:

The risks, plainly. Posting through the official API is allowed. Reading through Agent Reach is not: it uses my logged-in session, and LinkedIn's User Agreement bans that. I limit it to 40 reads a day, never in bulk, and the rule is to stop at the first warning. That lowers the risk. It does not remove it. One thing is still unconfirmed: whether my posting token can also like or comment. LinkedIn's docs contradict each other, so I test it once on my own post. As of today, Agent Reach's LinkedIn login is not connected and that test has not run.

What I would test next

Sources