Choosing between a shiny new personal AI agent and the ChatGPT or Claude setup you already use? Here’s the fast way to decide. I kept seeing headlines about Meta’s Muse, ChatGPT Dots, and xAI’s Grok Bot, and I couldn’t tell if they did anything new or were just better marketing. Then I found a clear breakdown from Jeff Su, a creator who makes AI productivity videos, and it sorted the whole thing out for me in about fifteen minutes.
Here’s what he explained, plus my take on how to choose.
The three levels of AI help
The author splits today’s AI tools into three groups. He uses one running example, a yearly health report, to show how each one is different.
- Chatbots (ChatGPT chat, Claude chat, Gemini chat): You upload the report and get a plain-English summary. It’s useful, but it’s a one-off. The chatbot can’t compare the report to your past results or your doctor’s notes, because it never saw them.
- DIY agents (ChatGPT Work, Codex, Claude Code, Gemini Spark): You give the agent a health folder with past reports, doctor’s notes, and your supplement list. When the new report arrives, it saves the file, compares it to the last five years, and updates a “memory” document so next year starts with everything it already knows.
- Personal agents (Muse, Dots, Grok Bot): These are the proactive version. The day before your checkup, the agent messages you first and offers a five-year summary of your health trends to bring along. You didn’t have to ask.
The creator calls the middle group “DIY” because you build the folders and files yourself. Personal agents come with that structure already in place.
What to weigh before you pick
To make a good choice, look at these six things:
- Setup effort: How much are you willing to build from scratch?
- Power: Do you need the most capable tools available right now?
- Control: Do you want to decide how your information is organized?
- Proactivity: Do you want the AI to reach out first, all day long?
- Lock-in: How easily could you take your data elsewhere?
- Privacy: How much personal data are you OK handing to a big company?
How the two compare
- 🛠️ DIY agents
- Pro: The most powerful option today. The expert’s own setup, called “Jeff OS,” holds more than 100 text files across 30+ folders, one for each part of his life.
- Pro: He didn’t create those files by hand. He tells the AI what he wants, and the agent writes, updates, and organizes everything.
- Pro: It’s just plain text, so it moves between Claude Code, ChatGPT Work, Gemini Spark, or open-source tools like Hermes and OpenClaw.
- Con: You have to know what to ask for. Nobody hands you a goal-setting interview or a starter checklist.
- Con: Words like “memory.md” and “claude.md” put a lot of people off. The author jokes that it sounds like NASA-level computer science.
- 🤖 Personal agents
- Pro: They work right out of the box. In Muse, the author opened “Goals,” picked health, and got interviewed about his sleep habits. Muse then saved the goal and built a bedtime checklist.
- Pro: They’re proactive. They keep checking all day on a cloud computer, even while your laptop is closed.
- Con: They’re still new and less capable than DIY setups.
- Con: Everyone uses the same standard system, so you can’t shape it much.
- Con: Lock-in. The more the agent knows about you, the harder it is to leave.
The part I loved most is when the creator opened Muse’s system files. They look almost the same as his DIY setup, with files like “agent.md” and “memory.md.” Both approaches work the same way under the hood. The AI saves what it learns in files and reads them when it needs to.
Real use cases worth stealing
The DIY examples really show what memory can do:
- Family planning: His friends-and-family memory file tracks birthdays, allergies, and past gifts. When he planned a surprise dinner for his brother, the AI checked the restaurant menu, flagged a severe shellfish allergy, and suggested a similar place nearby.
- Trip planning: One line, “Help me plan a trip to LA from May 20th to 30th,” sets off a lot of work. The agent finds Marriott hotels in his price range, checks visa rules for his passport, pulls LA friends from his Notion contacts list, builds a trip project the way he likes it, and flags calendar conflicts and deadlines.
- One small rule: He told the AI once that every reply should end with a “What’s next” section. It wrote that down and now follows it every time, which really cuts down his decision fatigue.
For personal agents, the strongest uses are about keeping watch. Think job alerts for a specific role and city, overdue invoice reminders, subscription renewal warnings, and price-drop alerts.
You might ask if a daily scheduled task would do the same job. The author’s answer is smart. If your dream job gets posted in the afternoon and closes before tomorrow’s run, a once-a-day task misses it. A personal agent checking all day catches it.
The privacy question
The expert’s view is blunt: privacy is a personal choice. He’s careful with Meta after Cambridge Analytica, so he uses Muse as if everything goes “straight to Zuckerberg’s personal inbox,” even though Meta says it can’t see your data. If privacy matters most to you, the fully private route is OpenClaw or Hermes running open-source models on your own computer. The trade-offs are a scary setup and weaker models.
My recommendation
- Go personal agent if you want the easiest start and don’t plan to build a system. Just accept the lock-in risk.
- Go DIY if you want the most power, like shaping tools around how you work, or want to keep your data portable.
I’m leaning DIY. The creator’s prediction makes it the safer bet: AI labs will likely add personal-agent features to their DIY tools, like OpenAI adding Dots to ChatGPT Work and Codex. Your folder of text files keeps its value either way.
How to get started
- Pick one area of your life, like travel, health, or family.
- Create an empty folder and point your DIY agent at it.
- Tell the AI your preferences in plain words and ask it to save them to a memory file.
- Add one reply rule, like ending every answer with “What’s next.”
- Connect the apps you already use, such as your calendar, email, or Notion.
- If you’re testing a personal agent, start with a low-stakes goal and keep sensitive data out.
The full video includes the actual Muse screens and a tour of Jeff OS, and it’s worth watching before you commit to either path.