Life when credits run out
I am currently at AltSpace in Dharamshala. This is my first experience with a coliving community, and the standards are set super high now.
Here I did my first hike, swum in a khad (cold cold cold), played board games and sports, and met people I’d remember for life.
Life’s good.
my work view
How I built a Twitter/X agent that posts 35x a week with no human in the loop (and get results)
This is probably early to share it as a playbook. But definitely a good time to publicly document version one.
I run a newsletter about coffee homebrewing for Indians with a friend.
This is a side gig for both of us, so it’s important we automate as many tasks as possible.
We noticed an increase in interest in coffee on Twitter/X and decided to repurpose our newsletter content for X.
So I built an agent (okay, workflow) that automatically repurposes and posts content on X.
My only role in this workflow is to approve the posts it generates by adding a “+” sign at the start of the posts.
The agent self-learns from engagement plus forms a thesis based on what posts I approve and disapprove. It documents its learnings and doubles down on what’s working. Twice a month, I review learnings.md.
I treat workflows and agents like employees. I set KPIs.
Primary KPI = Increase engagement and followers
Secondary KPI = Increase newsletter subscribers as a second-order effect
I was banking on discoverability and brand presence than the growth game.
On the 45th day, we got our first subscriber from our agent.
It’s the co-founder of a coffee roaster. He doesn’t match our reader ICP, but he is a potential partner. I emailed him, said hi and thanked him for subbing. We’re in touch now.
This is the point where I consider my AI experiment successful.
Not when I executed a complex workflow or the output looks dreamy. But when the results show.
I will continue to invest in our X account, because even the worst case is brand discoverability. No founder would mind that.
If you want to implement it yourself, this is all you need:
Tech:
Newsletter content as .md files locally
Repurposing by Claude Cowork as an automated task
I approve the tweets on VS Code
Pushed to GitHub. It picks the posts with “+” and posts them on X via X API
Another cowork action reads analytics via X API and creates its own learnings on what's working and what's not. Then creates next week's content based on the learnings
Cost:
X premium: $2.80 a month
X API: Less than $5 a month to post 100-140 posts and read analytics of all of them
Claude $20 a month
20 minutes a week to approve posts. 20 more minutes to iterate on the agent if I want to
Marketing resources shared in our community
How is the TMM community using AI?
This modern marketer replaced his video editing software with Claude
I run bi-weekly webinars.
Each webinar is ~30 minutes, recorded on Google Meet.
Every week, I had to trim each webinar into short, topic-based clips and post them on socials. But sitting through a full recording every week and manually editing it was tiring.
So I built “Webinar Clipper” with Claude that:
Reads the Gemini Notes transcript
Identifies distinct topics by timestamp
Splits the full recording into short clips
Write a LinkedIn post for each clip
I just run /clip and it’s done. No editing software or manual trimming. It’s not 100% accurate. But I don’t need surgical precision for these clips. So, it gets the work done and I’m happy with the output.
Submit your use cases and get featured in front of 1700+ readers