In a recent video, Alex Finn walks through a simple, low-cost way to run Clawdbot 24/7 on a cloud server—no Mac Mini required. On the surface, it’s a tutorial.
Underneath, it’s a clue about something bigger: the economics of AI are brutal, and the competitive pressure is rising.
When people start building “AI employees” that live in messaging apps and run nonstop, you’re looking at a future where compute costs, reliability, and distribution decide who wins. OpenAI helped ignite the boom. Now it has to survive the business reality of it—while Google’s Gemini keeps closing in.
Alex shows how to host Clawdbot on Amazon EC2 for roughly “subscription-like” monthly costs instead of buying new hardware. That matters because always-on AI workflows multiply usage—and usage multiplies compute costs.
Installing Claudebot on Amazon EC2: the quick, practical rundown
Alex also insists on one point: you don’t need to be a cloud expert to get Claudebot running. His EC2 setup is deliberately minimal, designed to get an always-on assistant online in under an hour.
Here’s a condensed, no-nonsense version of the process he demonstrates.
- Create an AWS account and open EC2. From the AWS console, search for EC2 and launch a new virtual server (“instance”).
- Choose a simple Linux base. Select Ubuntu as the operating system. This keeps compatibility high and documentation abundant.
- Select a flexible, mid-range instance. Alex recommends a flexible compute tier that typically lands around $15–$25 per month depending on usage—far cheaper than buying new hardware upfront.
- Allocate basic storage. Around 30 GB is enough to store logs, configs, and working files for early use cases.
- Create and save your SSH key pair. This file is your secure access to the server. Lose it, and you lose access.
- Open the required network port. Add a custom TCP rule (the Claudebot gateway port) so the assistant can communicate properly.
- Connect via the EC2 console. Use AWS’s built-in terminal to avoid local setup friction.
- Install Claudebot using the official command. Copy the install command from the Claudebot documentation, paste it into the terminal, and let it run.
- Choose your model provider. Plug in an existing ChatGPT or Claude subscription, or add an API key if you want usage-based billing.
- Connect your messaging channel. Telegram, WhatsApp, or Discord become the interface where your AI employee lives.
- Finish onboarding and “hatch” the bot. Name it, define its role and tone, and confirm your timezone—your assistant is now live 24/7.
The key insight is not the individual steps, but what they enable: a persistent AI agent that runs continuously, independent of your laptop, and whose cost profile looks more like a SaaS subscription than a hardware purchase.
That shift—away from devices and toward infrastructure—is exactly where the economics of AI start to matter.
What to do after installation: the first user cases Alex recommends
Alex doesn’t just show how to get Claudebot online. He also lays out a “starter pack” of workflows to make it immediately valuable. These aren’t flashy demos—they’re the kind of recurring, low-friction automations that turn an assistant into a habit.
1) Do a massive brain dump.
His first move is to feed the assistant rich context about who you are: your work, goals, preferences, relationships, routines.
The point is to make Claudebot useful without guessing, because persistent memory only helps if you give it good signal.
2) Set up a daily morning brief.
He suggests a scheduled message that hits you every morning with what matters: relevant news (based on your context), local weather, and actionable tasks that move your career or business forward.
The key is proactivity—your assistant should push value to you, not wait to be asked.
3) Start with one monitoring habit.
A practical example is email: ask it to summarize your inbox daily, flag what needs attention, and reduce cognitive load. In a cloud-hosted setup, this may require connectors and permissions, but the workflow is simple: “watch this stream and report back on a schedule.”
4) Ask the bot what it should automate next.
His favorite prompt is essentially: “Based on what you know about me, list five workflows you can do for me every day.”
That’s a way to surface “unknown unknowns”—use cases you wouldn’t think to request—then turn the best ones into scheduled tasks.
Why this matters for the bigger story: these workflows are the moment AI stops being “a chat tool” and becomes an operating layer. And once users run daily briefs, inbox scans, and recurring tasks, usage (and costs) become continuous—raising the stakes for providers like OpenAI.
What Alex’s Clawdbot setup reveals about AI economics?
Alex’s core point is simple: people are buying Mac Minis to run Clawdbot, but you can host it on the internet instead. The setup he demonstrates uses Amazon EC2—a virtual Linux server—so the assistant can run 24/7 without dedicated hardware.
That’s not just a budgeting trick. It’s an architectural shift. When your assistant is always running, it’s no longer “a chatbot.” It becomes infrastructure: persistent services, storage, connectors, permissions, logs, and a bill that scales with usage.
This is the uncomfortable part of the AI boom: the more useful AI becomes, the more frequently it runs. And the more frequently it runs, the more expensive it is to serve—especially when millions of people do it at the same time.
Always-on assistants change the cost curve
Alex frames Clawdbot as an “AI employee” that works 24/7. That’s the dream: proactive briefings, ongoing monitoring, scheduled tasks, and continuous memory across conversations.
But “24/7” has a price tag.
Even if your server is cheap, the intelligence layer—API calls to a model provider—costs money. Alex repeatedly emphasizes model choice as a cost/quality tradeoff, and even warns about reliability issues with cheaper options.
Zoom out and you see why OpenAI’s economics are tricky in consumer AI: if users expect constant availability, rich multimodal features, and low latency, costs don’t just rise—they rise nonlinearly.
