Multi-Agent Daily Operations for a One-Person Company
The author describes running one-person company workflows locally using multiple agents and scheduled tasks. Task: The author operates a one-person tech agency using four AI agents on OpenClaw, running locally on WSL2 with 25 systemd timers. Agents handle content generation, lead research, security scanning, and operations, utilizing Gemini 2.5 Flash free tier. The workflow includes quality gates where posts are self-reviewed and rewritten if scoring below 7/10. Human involvement is critical for meaningful client conversations and community engagement, while agents manage repetitive publishing and data syncing.; tools: Tool described in the source. Human role: Not stated in the source; limitations: This is a self-reported demonstration by a solo developer, not an independently verified production deployment. Key risks include accidental API billing if keys are created from billing-enabled GCP projects instead of AI Studio. The system has experienced bugs such as engagement loops burning rate limits and Telegram health check conflicts causing duplicate messages. Reliance on free tier limits means scaling requires architectural changes or paid plans..
Tools
Tool described in the source
Published on Agent Work Atlas:
Overview
The author describes running one-person company workflows locally using multiple agents and scheduled tasks.
Workflow
The author operates a one-person tech agency using four AI agents on OpenClaw, running locally on WSL2 with 25 systemd timers. Agents handle content generation, lead research, security scanning, and operations, utilizing Gemini 2.5 Flash free tier. The workflow includes quality gates where posts are self-reviewed and rewritten if scoring below 7/10. Human involvement is critical for meaningful client conversations and community engagement, while agents manage repetitive publishing and data syncing.
Reported result
The system generates approximately 105 requests per day, staying well within the 1,500 daily free tier limit, resulting in $0 monthly LLM costs. Reported outcomes include 27 automated Threads accounts with 12K+ followers and 3.3M+ views. However, the author admits the live agent dashboard visuals are currently CSS animations with data not yet fully piped from real logs. Revenue is derived from client projects and MindThread subscriptions, though specific MRR figures were not publicly disclosed.
Human role
Not stated in the source
Limitations
This is a self-reported demonstration by a solo developer, not an independently verified production deployment. Key risks include accidental API billing if keys are created from billing-enabled GCP projects instead of AI Studio. The system has experienced bugs such as engagement loops burning rate limits and Telegram health check conflicts causing duplicate messages. Reliance on free tier limits means scaling requires architectural changes or paid plans.
Try it
I want to pilot a multi-agent workflow for my solo business using OpenClaw and Gemini 2.5 Flash free tier. Please set up a local environment on WSL2 with systemd timers. Create two agents: one for drafting social media posts and one for scheduling. Implement a quality gate where the drafting agent self-reviews content and rewrites if the score is below 7/10. Use [replaceable inputs: my business context, target platforms, and source files for intelligence]. Ensure the agents use pre-computed local markdown files for context to minimize token usage. Before enabling any automated posting, run a dry generation for 3 days and provide me with the output for manual review. I will confirm the quality and accuracy before connecting any live APIs or scheduling actual posts. Measure success by comparing the time spent on manual drafting versus the automated pipeline, ensuring we stay under 100 requests per day.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.