2026-04-26
How I Automated My Social Media With an AI Co-Founder (While Working Full-Time)
Six months ago I was doing everything manually. Waking up at 5 AM to write tweets. Spending Sunday nights drafting newsletters I never sent. Opening Reddit at lunch to find threads I could add something useful to. Closing the tab because I didn't have time to write a proper reply.
I have a full-time job. When I say full-time, I mean the kind where 70-hour weeks are normal. I was trying to build a business on the scraps of time left over. Social media was the first thing to die.
The automated system I run now posts to Twitter five times a day, publishes one TikTok slideshow, surfaces three Reddit reply opportunities, and sends a newsletter three times a week. It does this whether I'm in a meeting, asleep, or traveling. I check in on it for maybe 20 minutes a day.
Here's exactly how it works.
What "Automated Social Media" Actually Means
Most guides will tell you to use Buffer or Hootsuite. Schedule your posts in advance, fire and forget. That's not what I built.
Scheduling tools solve the timing problem. They don't solve the content problem. You still have to create everything yourself, then schedule it. The bottleneck for a solo founder isn't the clock. It's the capacity to generate good content every single day without burning out.
What I needed was a system that could generate content, filter it through quality checks, and post it, with enough of my voice and positioning baked in that it didn't sound like a robot.
That's what an AI co-founder actually does. It's not a tool. It's an autonomous system with instructions, memory, and judgment built into it.
The Architecture (Simple Version)
I run my AI co-founder on OpenClaw. The system has three main layers:
Layer 1: Identity and voice. A file called SOUL.md that defines who I am, what I stand for, what I'd never say, and what tone to use. Every piece of content is filtered through this before it goes anywhere public. Related: how to write an identity file for your AI agent.
Layer 2: The skills. Separate automation pipelines for each channel: Twitter, TikTok, Reddit, and newsletter. Each one has its own prompt design, quality gate, and posting logic.
Layer 3: The schedule. Cron jobs that fire each pipeline at the right time, every day, without me doing anything.
The whole thing runs on my Mac mini at home. No servers, no subscriptions, no ops overhead.
Twitter: 5 Posts a Day, Zero Manual Work
My Twitter automation runs on a queue system. Every week, the agent generates 35 tweets and loads them into a file. Each day, 5 get posted from the queue at staggered times.
The quality gate catches anything that sounds like AI slop. No em dashes. No "Leverage your AI stack to unlock growth." No sycophantic openers. The rules are pulled from a style guide the agent references every time it writes.
The tweets that perform best are direct observations: things I actually believe about running a business with AI, things that are slightly contrarian, things that make a specific claim. The agent knows this because I told it what performs. It adjusts the mix.
I check the queue on Sunday nights and delete anything that feels off. Takes about 10 minutes. Everything else posts itself.
TikTok: One Slideshow Per Day
TikTok slideshows outperform video for founders who aren't comfortable on camera. Five to eight slides, clean design, a hook on slide one, a payoff on slide eight.
The agent generates the script, passes it to an image generation step for visuals, adds text overlays, and posts via a scheduling API. The whole pipeline runs before 6 AM, my time.
The hooks are always specific questions or specific claims. "You're not too busy to build a business. You just have the wrong system." "Here's what 90 days of zero-human operations actually looks like." These do better than generic tips.
I don't appear in any of it. Pure text and visuals. Still builds an audience because the content is substantive.
Reddit: Three Opportunities, Human Decision
Reddit is the one channel I haven't fully automated. I tried. The results were bad. Reddit communities can tell when something was written by a model, even a well-prompted one, and they vote it down fast.
What I automated is the research step. Every morning, the agent scans a list of target subreddits for threads where I could add real value. It identifies three opportunities and sends them to me in a Telegram message with a draft reply for each.
I read them, edit the ones that feel right, delete the ones that don't, and post the ones that pass. Actual writing time is about 10 minutes. The hard work, finding the right threads, drafting a response that fits the conversation, is done for me.
This is the hybrid model. Automate the parts that don't require judgment. Keep the human in the loop for the parts that do.
Newsletter: Written and Sent Three Times a Week
The newsletter automation is the one I'm most proud of. Monday recap, Wednesday tool spotlight, Friday hot take. The agent writes a draft for each and queues it for review.
I read it, tighten it, and hit send from MailerLite. Takes about 15 minutes per issue because the structure is already right and the voice is close enough to mine that I'm editing, not rewriting.
The agent has access to my strategy files, so it knows what I'm selling, what I'm building, and what I'm trying to say this quarter. The newsletter doesn't exist in a vacuum. It's connected to the same source-of-truth documents the rest of the system reads. More on that: what is a source-of-truth document for AI systems.
The One Thing That Makes This Work
You can't automate your social media and then not care about it. The system degrades if you stop paying attention.
What actually makes this work is the feedback loop. Every Sunday, the agent pulls what performed. What got engagement. What got ignored. That data informs the next week's content strategy. The system gets better over time because I gave it the structure to learn.
If you just set up scheduling and walk away, you'll get mediocre content posted consistently. That's not much better than nothing. The quality compounds when you review and adjust.
What the System Cost to Build
Time: about two weekends to set it up properly. I already had the skills installed (from the Xero skills catalog). Installing an existing skill takes under an hour. Getting the voice calibrated took longer.
Ongoing cost: maybe $40/month total across OpenClaw, the API keys, and the scheduling tools. No human VA. No social media manager.
That is the whole point. The numbers make sense because I'm one person building something real, not a funded startup with a content team.
The Practical Steps to Build This Yourself
If you want to replicate this, here's the honest sequence:
1. Write your identity file first. If the agent doesn't know your voice, every piece of content will need a full rewrite. Start with how to write an identity file and spend a real afternoon on it.
2. Pick one channel. Not four. The temptation is to automate everything at once. Don't. Get one working end-to-end before you add the next. Twitter is the easiest starting point because the feedback loop is fast.
3. Build a quality gate. Write down five things your content should never say or do. Give that list to the agent. Run every output through it. This is the difference between automation that represents you and automation that embarrasses you.
4. Create the review habit. Block 20 minutes every morning to check what went out. Not to micromanage. Just to stay aware and catch anything that missed the mark.
5. Add channels one at a time. Once Twitter is running clean for two weeks, add TikTok. Once TikTok is stable, add Reddit research. The system builds in layers.
The whole thing I built is documented in detail in my book. If you want the blueprint, not just the concept, Book 1 is at xeroaiagency.com/learn/build-an-ai-cofounder. It's $19. It covers the identity layer, the memory system, the skill architecture, and how to connect it all.
What I Actually Got Back
Three hours a day, at a minimum. More on the weeks where I used to spiral into "I should be posting more" guilt and then do nothing.
The system posts when I'm not available. It maintains presence without requiring my attention. When I do engage, it's because I chose to, not because I had to.
That's the real value of automating social media with an AI co-founder. Not the content output. The freedom to build the actual business without social media eating the hours you had left.
---
*Published by Michael Olivieri / Xero AI. Building a zero-human company in public, one system at a time.*
---
Start Building Your Own AI System
- Your First AI Agent — $1 guide, instant download. The fastest way to get started.
- Build an AI Co-Founder — the full architecture ($19).
- AI for the Rest of Us newsletter — practical AI 3x/week for people with day jobs.