I moved my LinkedIn comment drafting from n8n into Claude Cowork scheduled tasks. Here's what I learned.

Context: I'm building ALGA — an AI LinkedIn Growth Assistant.

The problem it solves: I can't grow LinkedIn consistently while running an agency and building hardware products. I can't hire a ghostwriter (cost). Generic AI sounds like everyone else. So I built something that actually knows me.

The original setup was an n8n workflow that would wake up every morning, grab posts from my target LinkedIn profiles via Unipile, draft comments, and save them to Airtable. It worked.

Then Anthropic dropped scheduled tasks in Claude Cowork. I spent ten minutes with it and moved the entire judgment layer out of n8n.

What the setup looks like now:

n8n still runs at 5am. It scrapes LinkedIn posts via Unipile, checks for duplicates, and uploads to Airtable. Data work — stays in n8n.

At 7am, a Claude scheduled task wakes up. It reads the scraped posts, loads my strategy table, my user table (full background, story, goals, skills), my companies table, my skills table. Then it drafts comments in my voice and presents them to me for review.

Two things that impressed me:

Claude skipped posts without me telling it to. A geography conference I can't attend. A giveaway with no signal for a builder. It used context to make judgment calls.
Creating the task took 3 minutes. I used /schedule inside a new task, spoke my instructions via Whisper, and Claude called the scheduling skill. Task created, 8am daily, Airtable connector auto-attached.

What's next: Turn the n8n scraper into a Claude scheduled task too. Then build the approval + posting flow as a third task. Full loop inside Claude Cowork.

Full build is on YouTube if you want to see the screen recording: https://youtu.be/i_68UvezeIg

Happy to answer questions on the Airtable schema or the scheduled task prompt structure — both are real, running right now.

0 comments