Category: AI & Product

With my Claude agents running on my MacBook Air, I thought I had a sweet setup. I could also use the Claude mobile app on my phone to remote in and ask it questions, with all of the context and files I have on my system.

This didn’t always work the way I thought it would:

• Sometimes the session would disconnect, and I could not see it on my phone.
• Sometimes the MacBook wasn’t plugged in, so I couldn’t reach it remotely.

Why an always-on agent

Enter Meta’s Muse agent and its idea of an always-on, 24/7 personal agent. I had heard about this before with OpenClaw, which you host yourself, and Grok Bot, which runs in the cloud. I was curious, so I tried to sign up. It wasn’t available in Canada at launch, but it now is.

I also thought about how I could recreate the same personal agent if I self-hosted. The one thing that really sets Muse apart is the connectors to other apps, and that’s what sold me on it.

My risk check before signing up

Before I signed up, I wanted to run a third-party risk assessment on it. These are the questions I asked:

1. Is the app safe, and is it a known developer? It’s from Meta, so it’s a known developer. Meta says a separate safety system checks its actions before they reach the internet, and it asks for my approval before sensitive actions like sending an email or making a purchase.
2. Is my data safe, and what privacy measures are in place? Meta says it runs in its own secure virtual machine. I can tell it to forget things, and I can disconnect any app whenever I want.
3. Can I stop my data being used for training? Yes. The setting “Help improve our AI models” is on by default, so I turned it off right away.
4. Is this company going to be around for the next 5 years? It’s a critical part of billions of users’ social media and communication, so I believe in its long-term outlook.
5. Is this a central focus for the company, or a side project? From what I understand, it’s a central focus for Meta and its way to take part in the AI race.
6. How would I leave? I can download my Muse data, and “Reset Muse” permanently deletes my chats, files and tasks. Meta notes that it may still remember things it learned from a deleted message, so a full reset is the cleaner exit.

It’s not perfect, and it changes with every update. That’s why I run this risk assessment again every few months, to make sure all the settings and data controls are intact. I also keep the risk low: I gave it a secondary email account and calendar first, to see how the connectors work. I don’t let it pay for things or send emails. I’m still being very cautious.

How I use it

After setting it up on my phone, I tried the website and, most importantly, the desktop app. Armed with my knowledge of using AI agents, I fed it some context about myself and my goals.

So far it’s been very helpful. I use it for its intended purpose, as a personal AI agent. It gives me a daily brief of:

• my emails
• my calendar
• things I’m watching, like market news, trips I’m planning and other random questions

One great example was when it flagged an important email that needed a time-sensitive response. I don’t check my email every hour, so I would have missed it.

It’s not all smooth, though. I haven’t figured out a way to build skills and give it deeper instructions for specific tasks, so I’ve had to correct it a few times.

The fact that it’s not tied to my laptop is really good. I also like that it has its own browser, and that there are connectors I can use and trust, like Asana.

Muse vs my Claude agents

I now think of Muse as my personal agent for trip tracking, important email reminders, the daily brief and research. I find its model, Muse Spark, good, but not Opus good. I use my Claude agents with Opus 5.5 for heavier tasks and more complex workflows.

In my simple model ranking of Low, Medium and High:

• Low: Sonnet
• Medium: Opus
• High: Fable

I would put Muse between Low and Medium.

My verdict

Would I recommend it? Yes, for casual users who want to try a personal agent for general use cases.

The same 6 risk questions work for any AI tool you’re thinking of trying.

I’m excited to see how Muse evolves and what it adds. I haven’t needed the paid version yet, as the tokens in the free tier are generous.

If you want to see how I set up my Claude agents, I wrote about my learnings in using AI agents.

I’ve been using AI for the last two to three years, but this year I went deep and started building my own agents. This is what I learned building one of them, an agent that helps me run this website.

Starting as a non-coder

As a non-coder, using Claude Code sounded scary. I haven’t coded since my programming classes in university, but I was determined to get the full benefits of AI for myself, so I set it up in the Claude desktop app.

I started with Claude Code, and I’ll use my WordPress agent as the example for the learnings and best practices below.

I created a dedicated folder for it and started with a simple prompt: “You are my website agent designed to help me manage my personal website.” Then it started asking me follow-up questions and we were on our way. No code required.

1. Pick the model and the effort

I experimented with different models and settled on Opus with High effort as the middle ground:

• Sonnet doesn’t work well as the orchestrator (the main agent that plans the work and hands parts of it to helper agents).
• Fable is overkill for what I need.
• For effort, Medium to High works well.

2. Give it a purpose first

In my experience, the most important thing for an agent is a clear purpose, and that’s the CLAUDE.md file. It’s a plain text file in the folder that the agent reads at the start of every session.

Once that was set, the next steps were:

1. Building the agent’s memory, so it keeps learning.
2. Teaching it skills and playbooks, for example how to run site operations and how to help me publish a post.

3. Manage the context window

As you keep chatting with the agent, the context window starts filling up. With Claude I get 1 million tokens, which sounds like a lot, but it eventually fills up.

My best practice is to keep a session state file (a short note of what’s in progress) and to regularly ask: “Update your skills and learnings so I can compact.” This way everything is saved, and I can compact the conversation when I choose.

4. Plan around the usage limits

My weekly usage limit resets on Sunday evenings, and you can only use so many tokens in a single 5-hour session.

So I now run scheduled routines to balance my usage through the day. These routines are automated sessions the agent starts on its own, for things like site maintenance.

I’m on the Max plan with 5x usage, and it’s the sweet spot for me with my current goals.

5. Clean up the instructions every month

Once a month, I do a prompt audit and ask the agent to improve and optimize itself. The agent and the way I use it keep evolving, so I want the instructions and skills to stay current, without garbage from old sessions or any conflicts.

6. When I get stuck, ask the agent

Often when I get stuck with the agent, I just ask the agent itself and have it give me three options.

What I’m experimenting with now

I’ve also been experimenting with spawning sub-agents that help complete tasks, with the main agent as the orchestrator for complex research tasks.

On top of that, I’ve been getting the main agent to talk to another agent I’ve been building as an executive assistant. These two are next level, and they still amaze me.

It keeps getting better

The more you use the agent, the more it learns, the more you can tweak it and the more it can help you. It’s a symbiotic relationship that keeps growing.

Overall, it’s been a fun experiment using an agent to help with my website and to keep growing my AI knowledge. It’s fun to talk to the agent in natural language, and to build in the guardrails and hard constraints that keep it safe. I wrote about the guardrails I set in How I used AI to improve my website.

How are you using your agents, and do you have any best practices? Let me know.

This summer I let an AI agent work on my website, and it did more in a few sittings than I had done in years.

I am always working on improving myself, with new personal and professional goals, and I wanted my website to reflect that.

I have been experimenting with Claude Code agents a lot this summer on personal projects, so I built one just for this site. It was a fun experiment to see what’s possible.

From a product perspective, the what and the why were clear. I wanted a reader of the site to be able to:

1. See clearly what the site is about
2. Easily find information about me
3. Connect with me through the contact form or social media

I then broke my goals down further:

• Speed: This was a big one. I wanted the site to be fast and snappy. It’s a simple personal blog, so it should not be bogged down.
• Security: This was also important from a risk management perspective. My degree is in networking and IT security, so I wanted the site to be up to date, with all of the security measures in place and updates turned on.
• Layout and design: I wanted the site brought up to modern design standards. I have been blogging here since 2013, so I wanted to give it a revamp for the AI era with a clean and simple design.

Getting the agent connected

With the goals defined, I got to work and set up a Claude agent on my computer. I wrote my goals into its CLAUDE.md file. My first hurdle was getting the agent to connect to the site on my behalf. MCP is the standard way AI agents connect to other apps. There were some paid MCP connectors, but I didn’t want to pay for another subscription. Instead I used the free MCP Adapter plugin for WordPress and had Claude build the custom tools it needed to read and edit my posts and pages.

Speed and security

Then the fun began. The agent ran a full speed and security audit. It went through every plugin and setting and recommended what to change. I reviewed each recommendation carefully, and I taught Claude a skill to take a backup before any big change. Here is what changed:

• 10 old plugins removed
• 3 unsupported plugins replaced with modern ones (email, two-factor login and the privacy policy)
• PHP, the language WordPress runs on, upgraded from version 7.4 to 8.3
• The database cleaned out from 103 MB to 34 MB
• A Google PageSpeed score of 99 to 100

This could easily have taken days of my time. It was all done in a few evening sessions.

What I kept in my own hands

Giving an AI agent access to my site was also a risk decision, so I set some rules from day one:

• It never sees or types my passwords. I log in myself.
• Nothing goes live without my OK, and it shows me a preview first.
• It takes a backup before any big change.
• Anything permanent, like deleting old data, I click myself.

These are the same controls I would want on any AI agent at work.

Layout and design

This was a good start. The site was faster and I felt good knowing I had made it better. However, the layout and design were still lacking. It was a decent layout on the base theme, but it looked dated. I asked my agent to review the site as a web designer, a UI expert and a random reader, on both mobile and desktop. The suggestions came in, I tweaked them a bit, and then I watched it make the changes for me by controlling my browser. I was astonished! Each round made the design cleaner and easier to use.

What I learned

1. Write your goals down first. The agent was only as good as the goals I gave it.
2. Review every change. The recommendations were good, but I still tweaked a lot of them.
3. Teach it once. Every skill I taught the agent is there for the next time.

Now I feel confident in my site for 2026. It’s ready for blogging again and represents me well online. Plus, all of the effort I put into my Claude agent and its skills means its knowledge will keep growing, and I can keep improving the site.

If you have tried something similar with your own site, I would love to hear how it went.