Claude Code, Agent Factory, RobOS
Three ways to turn your Claude subscription into a system that ships. A direct comparison, with numbers from real production use, so you know which one fits your problem.
Why there are three different answers to the same problem
Companies buy AI tools one at a time. One for writing. One for video. One for research. Then they find themselves with four subscriptions and zero system.
The real problem is not that Claude (or any other model) is not good enough. The problem is that Claude is an engine. An engine on its own produces nothing. You need a machine built around it. That is where the split begins: what kind of machine.
I have built two different machines for two different needs, both on top of the same engine. Agent Factory is the machine that ships while you sleep, articles, videos, social posts, autonomously. RobOS is the workshop that makes you several times more productive while you work, with calibrated skills, persistent brand voice, visible cost per task.
This page is the map. It helps you figure out where you stand and what to pick.
The shared mechanism: Claude with muscle
Both products exploit the same finding: your Claude subscription gives you access to Opus and Sonnet through Claude Code. And through Claude Code, your programs can consume the subscription quota instead of burning money on the API key.
The internet already has dozens of wrappers promising exactly this: AI editors like Cursor, code agents, writing assistants. Almost all of them work on the same principle, with one detail that never shows in the ad: they first consume the quota included in your subscription, then automatically enable extra usage and start billing per token at standard API rates. The more you use, the more you pay. At serious volume you can hit thousands of dollars a month without noticing.
Agent Factory and RobOS deliberately take the opposite path: they stay strictly on the Claude subscription, with no extra usage enabled. When the limit is hit, the agent waits for the period to reset, no per-token invoice. The cost stays fixed, regardless of how much work they produce inside.
The three pieces
Each one with a clear role. They do not compete, they stack.
THE ENGINE
Claude Code
Anthropic’s CLI. Generic, powerful, opinion-free. It already has memory, skills, scheduling, hooks. You tell it what to do, it does it.
- Runs on laptop, desktop, or server
- Access to Opus, Sonnet, Haiku
- Pro subscription from $22/month
- For any generic power user
AUTONOMOUS
Agent Factory
Course plus stack that installs on an Ubuntu VPS a system of four AI agents working 24/7. You send a Slack message, the agents execute.
- 16 modules N0-N15 installed step by step
- 4 agents: dispatcher, writer, video, security
- WordPress, video pipeline, social media
- $0.02 per article, $0.11 per video
ASSISTED
RobOS
An operating system for your daily work with Claude. Dashboard, 27 pre-built skills, persistent brand context, cost tracking, quality scoring. Runs on your laptop.
- 27 calibrated skills (marketing, content, research)
- 5-level memory mapped to real files
- Multi-client with full isolation
- Your data stays on your machine
Comparison table across 14 dimensions
Read horizontally. See where each product is strong and where it does not apply.
| Dimension | Claude Code | Agent Factory | RobOS |
|---|---|---|---|
| Product type | Anthropic CLI | Course + VPS install | Commercial product (dashboard + skills) |
| Mode of operation | Manual, session by session | Autonomous 24/7 (cron + heartbeat) | Assisted (you work, it structures) |
| Where it runs | Your laptop or machine | Dedicated VPS | Laptop (local data, SQLite) |
| Main interface | Terminal / IDE | Slack DM (remote control) | Web dashboard with 8 sections |
| Memory | Native (CLAUDE.md, memory dir) | Seed memory per agent | 5 levels mapped to real files |
| Skills | Native (generic, opinion-free) | Hardcoded per agent | 27 pre-built skills + analytics |
| Scheduling | Native (cron, /schedule) | System-level crontab | Visual dashboard with cost tracking |
| Brand context | No | Fixed for the VPS agents | 5 cross-platform files (voice, audience, positioning) |
| Multi-client | No | No (1 VPS = 1 stack) | Yes, full isolation |
| Main output | Whatever you type | WP articles, TikTok video, posts (autonomous) | Copy, SEO, email, research (assisted) |
| Cost tracking | No (global Anthropic billing) | RoboMetrics N14 (per session) | Per skill / task / client |
| Quality scoring | No | No | On every output |
| When it is active | When you open the terminal | Non-stop, including overnight | When you work |
| Starting price | $22/month | $649 one-time | $97 one-time |
| Target user | Generic power user | Non-technical founder who wants „set and forget” | Freelancer / agency working with AI daily |
What this looks like in practice
I do not sell ideas. Both products run in production, including on my own internal system and across the cohort installs.
- My internal system: 18 active agents in RoboMarketing, across several stacks, handling client work at full tilt.
- Agent Factory cohort 1: 10 active users, each with their own 4-agent stack. All stacks have been producing articles and videos for months.
- RobOS: used daily on my own projects and on client work. The dashboard below shows the per-skill distribution over the last 30 days.
Per-Skill Breakdown
| Skill | Runs | Quality avg | Avg duration |
|---|---|---|---|
| tool-humanizer | 184 | 8.7 | 4.2s |
| content-copywriting | 73 | 8.6 | 31s |
| content-repurpose | 51 | 8.5 | 22s |
| research-trending | 42 | 7.9 | 34s |
| content-blog-post | 38 | 8.2 | 67s |
| sys-daily-plan | 27 | 8.9 | 14s |
| research-competitors | 18 | 8.1 | 89s |
| sys-session-close | 16 | 8.5 | 12s |
| sys-audit | 12 | 8.3 | 45s |
| brand-voice | 8 | 9.2 | 142s |
| + 18 other skills, 487 runs total | |||
RobOS dashboard, last 30 days. Real cost $0 because everything runs on the Claude subscription. API equivalent: roughly $1,847. The per-skill distribution reflects real work, humanizer is the most used (applied on every published article), then copywriting and repurposing.
A workday with all three
This is what my day looks like when I run all three systems in parallel.
This is not theory. This is the flow I have been running for months on RoboMarketing and on client projects.
Who is telling you this
I am Adrian Ulmeanu, founder of RoboMarketing. Over 30 years in IT and automation infrastructure. I build and use all three systems described above every day.
I wrote Visual Automation Atlas (the book that explains automation patterns through diagrams), I built Agent Factory as an alternative to „hire another copywriter”, and I packaged RobOS as a commercial product for professionals who were hitting the same frustration as me: Claude is powerful, but every single time you have to re-explain who you are.
RoboMarketing is the agency through which I deliver these systems to end clients. If you want to talk directly or see what would fit your business, message me on WhatsApp, email office@robomarketing.ro, or visit robomarketing.ro.
Which one fits you?
Two questions. Thirty seconds.
