Adding an agent to my personal blog repo: Is it worth it?

Over the past few weeks, I’ve been experimenting with Hermes, an AI agent, to test the promise we hear everywhere: Can personal AI agents actually make us more productive?

It started with a simple idea: What if I gave an autonomous agent access to non-critical work folders? Eventually, I decided to give it access to the repository for my entire personal website.

I gave it permission only to propose and commit changes to its own branches. Never push to production.

Today, I have a local agent powered by my LLM of choice that interacts with me via Telegram and commits code based on our chats. It genuinely feels like a personal assistant: it created a content calendar, integrated with my docs, and scraped the web for ideas from sites I follow.

The real question is: Is it actually worth it?

A Controlled Experiment

From the beginning, control was my top priority. I needed to isolate what the agent could touch.

Initially, I ran it locally using Docker, but every time I closed my laptop, Hermes went to sleep, defeating its purpose. To fix this while keeping things secure, I established a few strict boundaries:

• Dedicated Virtual Machine: Hosted in the cloud (DigitalOcean) so it runs continuously without local access to my private machine.
• Short-Lived GitHub Tokens: Scoped repository access with quick expiration in case of a token leak.
• Sandboxed Google Drive Access: Connected strictly to an account holding non-sensitive, low-priority documents.

Infrastructure Costs for an Agent That Never Sleeps

Running an agent 24/7 quickly revealed the infrastructure costs behind the setup.

A bottom-tier cloud instance wasn't enough; memory and compute requirements kept pushing me into larger, more expensive droplets. While hosting in the cloud meant I didn't need to leave my MacBook running all day, the monthly hosting bill grew surprisingly fast for a single personal project.

Still, the setup has real advantages:

• 24/7 Availability: Background workflows run continuously regardless of whether my laptop is open or I’m traveling.
• Persistent Context: Hermes maintains awareness of my tech stack, site architecture, and ongoing projects across conversations.
• Frictionless Mobile Access: I can delegate research tasks or discuss site tweaks directly via Telegram from anywhere.

Safe Deployments: Commits, Not Pushes

Hermes cannot push directly to production. Instead, it spins up isolated branches, tests layout tweaks, and drafts features derived from our Telegram chats.

From there, the workflow is deliberate:

Hermes opens a branch and makes commits based on our discussion.

I pull the branch down locally on my Mac to review the changes.

If everything checks out, I merge and push to production myself.

This keeps a strict human-in-the-loop safeguard and makes rollbacks trivial if an experiment misses the mark.

Automating the Process, Not the Writing

When I first gave Hermes access to the codebase, I asked it to draft content. It did the job, but writing articles isn't what I actually want out of this.

Instead, I use it as an automated research engine:

• Performing weekly scans across industry blogs and web properties.
• Maintaining an idea backlog synced to Google Sheets.
• Reviewing structure and notes across Google Doc drafts.

Conclusion

So far, I haven’t seen major net productivity gains. In fact, managing the VM, reviewing branches, and refining prompts has created almost as much work as it saves.

There are simpler, cheaper ways to achieve these workflows today. But as an exploration of what self-hosted AI assistants are capable of, it’s been an interesting experience.

The Stack

• Site & CMS: Astro, Tailwind CSS, TypeScript, Sanity CMS
• Hosting & CI/CD: Netlify
• AI & Automation Agent: Hermes (Nous Research), Google Gemini
• Code Review & Local Tooling: Antigravity
• Infrastructure: Ubuntu Droplet on DigitalOcean, Docker
• Research & Integrations: blogwatcher-cli, Google Sheets API, Google Search Console API

If you want to set up your own persistent Hermes instance, check out the DigitalOcean tutorial: How to Run Hermes Agent on a Virtual Machine.