Newsletter · Issue 03

The Human in the Loop

Practical AI for work, family, and everything happening in between.

Issue 03 October 1, 2026 About a 6-minute read

I have two new coworkers.

One is a mini black schnauzer named Olive. The other is Lola, a mini white schnauzer who is also my Meta Muse assistant, and who floats over my screen like she has a calendar full of opinions. They are fashioned after my two real coworkers, currently asleep under my desk.

They are adorable. They are also a useful reminder that an AI assistant can feel more approachable when it has a little personality and a face. The face is the fun part. The real value is still what you ask it to help you do.

So this week:


A few things worth knowing

The AI updates are starting to sound less like “ask me anything”

Tool · Anthropic

Claude Sonnet 5.5, built for the everyday work

Anthropic introduced Claude Sonnet 5.5 this week, describing it as faster and less costly for most work than its predecessor, with particular strengths in everyday tasks and polished documents, slides, and spreadsheets. That is the part I would notice if you use Claude for the work that is important but not quite important enough to clear your afternoon. Read Anthropic's Claude Sonnet 5.5 announcement.

Tool · OpenAI

GPT-6.1 Sol, strong work performance at a lower cost

OpenAI also introduced GPT-6.1 Sol this week. The practical headline is its stated balance of strong work performance and lower cost than its top-tier model. OpenAI says it is available in ChatGPT Work and Codex for eligible plans. For most of us, the useful test is not a benchmark: give it a real document, a messy set of notes, or a multi-step task and see whether it produces a better first draft with less back-and-forth. Read OpenAI's GPT-6.1 Sol announcement.

The takeaway

Model names will keep changing. You do not need to change your entire routine every time. Try the new option on one familiar task, compare the result with what you already use, and keep the one that saves you time without making you do cleanup for an hour.


If Copilot is your workhorse

Give your Copilot a job description, not just a question

You can give a custom Copilot agent a name and personality through its instructions, which makes it feel more personal. You still choose or mention the agent to use it; the name is not a voice command. In a regular Copilot chat, tell it what kind of help you want and what boundaries to respect.

Prompt · Meeting-prep partner
Be my meeting-prep partner. Help me get clear on the purpose, the people, the decisions, and the questions I should bring. Separate confirmed facts from assumptions. Don't send messages, schedule anything, or make commitments for me.

If you can add reference material in your work environment, give it the meeting brief or documents you are allowed to share. If you cannot, paste in the approved details. A good working style travels across tools; the pet is optional.


Try this week

Make yourself a tiny desktop companion

Meet the coworkers

Olive is my ChatGPT desktop pet. Lola is my Meta Muse assistant. They work differently, and your results may not look like mine. What I love is how much friendlier the work feels with a little personality sitting in the corner.

And once you have your assistant nearby, try the same “give it a role and boundaries” move for:

If you use ChatGPT's desktop app

On Mac or Windows, the current Pets feature lets you choose or create an animated companion that can float over other windows. Its controls can start a chat and show activity while you keep working. It is a delightful little “what is my AI doing now?” indicator. And to be precise: changing the pet changes how it looks, not how ChatGPT thinks or behaves.

Try this:

  1. Update and open the ChatGPT desktop app. Go to Settings > Pets, then choose Create pet. Pets may be restricted by your organization or may not have reached your account yet.
  2. When the pet-creation chat opens, paste the prompt below. Make it yours: swap the animal, colors, accessories, or mood. Upload reference photos of your actual pet or idea for context.
  3. When it finishes, return to Settings > Pets, select Refresh, and choose your new companion. Drag it where you like.
  4. Use Option + Space on Mac or Windows + Alt + P on Windows to open the floating controls and Quick Chat. You can type or use voice, then check the activity controls for progress.
Prompt · Desktop companion
Create a small, animated desktop companion named [NAME]. It is a [ANIMAL OR CHARACTER] with [DISTINCTIVE DETAILS: color, clothing, accessory, or expression]. Make it charming, polished, and easy to recognize at a small size. Give it a calm resting look and clear, lively poses that make it feel present while I work. Keep the design simple, warm, and professional enough to sit on a work desktop. See my attached photos for reference.

Mine is Olive, a black mini schnauzer with the sort of face that suggests she has just heard the word “walk” or “treat.” (Cue tail wag.) Check your company's rules before sharing work screens or files. See OpenAI's current Pets setup guide for the steps, shortcuts, and availability notes.

One small gotcha

Quick Chats started from the pet controls sit outside a Project. If you need your Project's files or context, start the chat from inside that Project.


Try something new

Meta Muse

Meta says you can personalize Muse's name, personality and tone, communication style, memories, avatar, and tagline. You can ask for those changes in chat or edit them manually. I tried the photo approach with Lola: I attached a picture of my real dog and asked Muse to use it as the reference for her avatar. That was my experience in my account, so the exact controls may vary.

Give your Muse a look and a voice

  1. Open Meta Muse and start a conversation. Attach a clear photo of your pup if you want to use one as the reference for the character.
  2. Tell Muse what you want to personalize: the assistant's name, the features that make your dog recognizable, its tone, and how you want it to communicate.
  3. Review the result. Ask for a revision if the avatar or style is off, or adjust the identity settings manually. Meta says these customizations can carry across conversations.
Prompt · Into Muse
Use the attached photo of my dog as the visual reference for my Muse character. Her name is [DOG NAME]. Keep the features that make her recognizable: [DISTINCTIVE DETAILS]. Please call this assistant [ASSISTANT NAME], use a [TONE] tone, and communicate in [STYLE, such as concise answers or friendly step-by-step explanations]. Update the avatar and these personalization settings only. Do not connect apps, change permissions, or take actions.

For the current customization options, see Meta's Help Center: Customizing Muse's personality and memories.

Coming up

This week, start with the character. Next week, I'll take a closer look at what you can expect from Meta Muse and how to think about its broader capabilities.


Before you go

Introduce your coworkers

I did not expect to spend part of my workday introducing two AI assistants to each other like they were arriving at a neighborhood book club. And yet, here we are.

If you make one, give it a name. Give it a look. Let it sit in the corner while you work. Just remember: the cartoon eyebrows are decoration. Your judgment is still the executive function in the room.

Show us yours

If you do make one, share your Loom or a photo and we will showcase your work in the newsletter in two weeks.

Submit your assistant »

One thing I'd tell a friend

You don't need to become an AI person

You don't need to become an AI person to make AI more useful. Start with one small job, tell it how you want help, and be clear about what it should leave in your hands. If a tiny dog makes you smile while you do it, that counts as a perfectly respectable user experience.

I'll keep sorting through the rest with you.

Michelle Hamilton
AI Adoption & Change Management Practice Lead · AnswerRocket
Back to AI × CRE