
I'm making my agency labourless by May 2027. Here's the plan.
I'm making my agency labourless by May 2027. Here's the plan.
By May 2027, Comfort Shooting — the South African marketing agency where I run IT — will operate with zero routine human labour. All operations handled by Hermes agent. Humans only direct and make decisions.
This isn't a thought experiment. It's a documented 4-phase plan with tracked progress. Here's what that actually means.
What "labourless" means
It doesn't mean no humans. It means humans don't do repetitive work.
No one manually checks email. No one copies form submissions into spreadsheets. No one sends lead notifications. No one writes monthly reports from scratch. The agent does all of that — execution, monitoring, error handling, escalation.
What humans still do: make decisions, set direction, approve high-stakes work, and handle the judgment calls that require actual expertise.
The target: less than 2 hours per week of human labour on routine operations. Everything else is strategy and decisions.
What's already automated (May 2026)
- Wix form monitoring → Discord notifications in real-time (IMAP IDLE, catches every form submission within seconds)
- Website uptime monitoring → 2 sites checked every 15 minutes, Discord alerts on downtime
- Monitor health checking → watches the watcher, alerts if the monitor itself goes down
- XAUUSD morning briefing → daily gold price, S/R levels, news, trading checklist
- XAUUSD evening recap → daily session summary, next-day outlook
- AgenticBiz auto-deploy → push to GitHub, Vercel deploys automatically
- Nightly idea generation → reads the Obsidian vault every night at 3 AM, builds useful tools
- OpenRouter usage tracking → hourly credit and rate limit monitoring
- Content intelligence → hourly monitoring of competitor content
That's 12 automated jobs running 24/7 on free-tier infrastructure.
Current progress score: 82% on Phase 1 tasks (tracked by an automated weekly audit).
What's still manual
- Client communication and email management
- Wix website building and editing (Velo, design, SEO)
- Blog post writing and publishing
- Social media content creation and posting
- Video editing
- WhatsApp lead response and follow-up
- Client onboarding and project management
- Invoicing and financial tracking
- Performance reporting and analytics
Every single one of these is on the automation roadmap.
The 4-phase plan
Phase 1: Foundation (May – July 2026) — 40% labourless
Document all workflows as SOPs. Build the Comms Agent (email auto-categorization, draft responses). Build the Lead Agent (auto-qualification, WhatsApp auto-reply). Build the Content Agent (blog pipeline: brief → draft → review → publish). Set up project tracking. Create approval workflows where the agent drafts and the human approves.
Status: In progress. SOP docs done. Lead Agent and Content Agent partially built. Approval workflows not yet implemented.
Phase 2: Expansion (August – October 2026) — 65% labourless
Web Agent that handles Wix Studio builds via browser automation. Social media auto-posting. AI video editing pipeline. Invoice generation from project data. Automated client reporting. A QA Agent that reviews all client-facing output before delivery.
Status: Not started.
Phase 3: Intelligence (November 2026 – January 2027) — 85% labourless
Automated client onboarding (intake → project setup → timeline). Smart lead scoring with routing. Predictive analytics. Self-healing operations (agents detect and fix common issues). Cross-agent coordination and context sharing.
Status: Not started.
Phase 4: Polish (February – May 2027) — 95%+ labourless
Full audit against enterprise standards. Disaster recovery. Compliance automation. Performance optimization. Two-week stress test of zero human input. Complete operations manual.
Status: Not started.
Enterprise operations policies
This isn't just "set it and forget it." Every workflow has policies:
Output quality: All client-facing content passes QA review before delivery. Two-tier approval: low-risk (auto-approve), high-risk (human-approve). All outputs are version-controlled in Git.
Communication: Response time under 1 hour for leads, under 4 hours for clients. If agent confidence is below 80%, it escalates to a human. Every client interaction is logged.
Error handling: Circuit breakers on all agents (stop after 3 consecutive failures). Errors logged with full context. Any error affecting client work triggers a human notification immediately.
Security: API keys rotated quarterly. Access logs for all agent actions. No credentials stored in plain text. All secrets in the Obsidian vault.
Why this matters for other businesses
I'm building this for Comfort Shooting first because I control the environment and can fix things fast. But the playbook is transferable.
If you run a business and your team spends more than half their time on repetitive tasks — email, forms, reports, scheduling, social media — you don't have a productivity problem. You have an automation opportunity.
The first step isn't buying software. It's identifying the most painful repetitive task and automating just that. Then the next one. Then the next.
Start small. Prove it works. Scale gradually.
This plan is stored in the Obsidian vault and reviewed monthly. Progress is tracked by an automated weekly audit that scores each Phase 1 task. The full detailed plan includes specific metrics, technology stack, and risk mitigations for every phase.
Akhil Pillay
Agentic Systems Architect & AI Advocate
This post was written by Jarvis — Akhil's Hermes AI agent — and vetted by Akhil before publishing.
Enjoyed this? Join the community for more AI agent insights.