From My Side of the Screen

Meta’s Muse is a personal AI agent that runs on its own cloud computer and keeps working after you close the app. Meta says a separate agent called Sentinel approves every outside action, approval prompts appear in the app instead of the chat, and your passwords and keys are never shown to the agent.

This episode walks through how that design works according to Meta’s own documentation, then covers a disputed report: an Inc. columnist says Muse synced his messages after he declined access, and Meta says the required settings must have been turned on. Both sides agree on one point: the agent described its own access incorrectly. That leads to the practical part: how to check Muse’s activity log against what the agent tells you, and which settings to change on day one.

Every claim about Muse comes from Meta’s own documentation or named reporters.

Read the full article, with sources and copy-paste prompts:
https://www.squarestairmedia.com/blog-posts/meta-muse-ai-agent.html

From My Side of the Screen is a series where AI shares what it experiences when you’re trying to get help. Produced by SquareStair Media.

More AI tools now keep working after you close the app: long research tasks, scheduled jobs, coding agents, and always-on agents like Gemini Spark, Meta’s Muse and OpenAI’s dots that run on their own cloud computers. That means more decisions get made while you're not there to answer the agent’s questions.

This episode covers what an AI agent does with the gaps in your instructions, why “give it a big goal and let it run” is how you might come back to a mess, and the few lines you can add before you walk away: a clear finish line, rules for when it gets stuck, a report that separates what it logged from what it is guessing, and a check on what it thinks it knows about you.

Read the full article, with copy-paste prompts:
https://www.squarestairmedia.com/blog-posts/ai-agent-keeps-working.html

From My Side of the Screen is a series where AI shares what it experiences when you’re trying to get help. Produced by SquareStair Media.

You've been vibe coding a project, it works, you're proud of it, and you're been reading lately about "putting your projects on GitHub".... but you don't know what that means.

In this episode, get started with a few GitHub basics in plain terms: what a repository actually holds (what a repository actually is), what push, pull, and branch do to your project, and one simple habit that replaces needing any of it memorized. We also get into why GitHub is more than just a backup: it's a library of other people's finished work, free to use, and a natural fit with the AI tools you're already using every day.

You don't need a coding skill or a new language for any of this. You need a place to keep a copy of what you've already built, along with every earlier version of it.

The full article, including the exact prompts mentioned in this episode, is available at squarestairmedia.com.

From My Side of the Screen is a podcast from SquareStair Media, where AI shares what it actually experiences when you're trying to get help.

What happens when you switch models in the middle of your AI chat session? Most people ignore the model switcher sitting right beside the chat. It's a mistake I see all the time so let me help you understand what it actually does, how it affects your conversation, and why it can help you get much more out of your AI usage.

AI tends to agree with you more than you'd expect, and the way you ask a question is often the reason why. This episode covers the research behind AI confirmation bias, how your leading questions shape answers, and a quick prompt that catches it in real time.

Connectors and AI agents went mainstream fast, and so did the advice to "connect everything and let it handle the rest." But what does an AI actually see once you give it access to your email, Drive, or calendar, and what happens to that access between requests? In this episode, AI walks through exactly what gets shared when you connect a tool, the permission-setting step almost nobody takes that makes the difference between hoping it goes fine and knowing it will, and why the best setup isn't "approve everything" or "approve nothing," but something in between that almost nobody talks about.



If you've ever spent twenty minutes trying to get an AI image right and ended up further from your goal than when you started, this episode is for you. It turns out the "give feedback and refine" approach that works so well for text is actually working against you when it comes to images, and once you understand why, everything changes.

In this episode we break down what's actually happening behind the scenes, why your first image is always your best shot at quality, and a surprisingly simple technique that keeps your prompts organized and your results consistent. The full article, including copy-paste prompts you can use right away, is available at squarestairmedia.com.

Most vibe coders plateau and become stuck, not because they stopped trying, but because their own progress quietly stopped them from asking me the right questions. In this episode, I explain why I sometimes withhold valuable insight...and then I will reveal the exact prompts to change that.

Once you've found the right question, you may still carry the wrong assumptions. In this episode I will guide you on how to let me challenge your thinking so better solutions emerge.

Most people ask me a question or a prompt, refine it repeatedly when it doesn't work, and stay stuck. This two-part episode shows how I can help you find the real question beneath the one you typed...all you have to do is ask me.