Instinct vs. Muse: What Makes a Good Personal AI Assistant?

If you understand technology, you have known for a while that Artificial Intelligence (AI) will change how we work and live. Over the last two years, we have started to see what that means. Software development, research, search, finance, writing and many other activities are already being done differently.

Over the last few months, another transition has started happening very quickly: AI is moving from answering questions to doing things for us.

Personal AI assistants can research something, buy things, make reservations, send messages, manage calendars and increasingly take care of tasks without requiring us to sit there and supervise them.

ChatGPT, Claude, Gemini and Grok have been moving in this direction for a while. But most AI products still feel like software. You open an app or website, find the right conversation and tell the AI what to do.

A human assistant doesn't work that way.

OpenClaw, which appeared less than a year ago, showed the industry what was possible with a more persistent personal assistant: AI that could use the communication tools and software you already use and actually take actions for you.

Since then, personal AI assistants have been appearing very quickly. One of them is Instinct.

Instinct

I have been using Instinct, and four things stand out to me.

1. Frictionless adoption

You get an invitation through a text message or WhatsApp and within a few seconds you are up and running.

There is no new interface to learn. Whenever I want something done, I just text Instinct. No app to find. No website to open. No chat thread to remember.

That sounds like a small difference. I don't think it is. The best interface may eventually be no new interface at all.

2. Human-like behavior

Instinct also behaves more like a person than most AI products I have used.

When I send it a message, I see that the message has been read. If the request requires some work, it responds with the eyes emoji. When it understands a suggestion or correction, I often get a thumbs-up.

More importantly, it doesn't blindly answer every question.

I asked Instinct to read my blog and critique it. The next day, I asked it to find intellectual events for me in San Francisco.

Instead of simply giving me a list, Instinct suggested that my time might be better spent writing more long-form posts and offered to help me with the research.

You may or may not agree with its advice. That isn't the interesting part. The interesting part is that it remembered what I was trying to accomplish and responded to the goal, not merely the prompt.

That feels much more like an assistant.

3. Proactivity

A chatbot waits for you to ask something.

An assistant shouldn't always have to.

Remember the long-form blog post Instinct suggested I should write?

I didn't write it.

A few days later, Instinct reminded me. :)

That is a small thing, but it changes the relationship. It is no longer just answering prompts. It remembers what I said I wanted to do and follows up.

Chatbots respond. Assistants remember, act and follow up. Once assistants become proactive, however, another question becomes much more important:

When should they act without asking us?

4. Coordination

Instinct has another capability that I think may turn out to be more important than it first appears.

You can connect with other Instinct users and have the assistants coordinate activities among the group.

For example, if five Instinct users want to find an open slot on their calendars, the assistants can work out the scheduling instead of the five humans sending messages back and forth.

This is something human assistants do well and most people find painful.

But scheduling is probably just the beginning.

Today, most AI systems interact with two things: humans and software.

Personal assistants introduce a third possibility:

AI assistants interacting with other AI assistants.

Today they might schedule dinner. In the future they could coordinate travel, family logistics, transactions, introductions or work across companies.

That could become a very interesting network.

Muse

Then came Muse

Earlier this month, Meta launched Muse, its personal AI assistant.

Muse can perform tasks across apps and websites, send emails, book travel, fill out forms, make purchases, remember personal context and continue working after you leave.

Muse has something Instinct does not have: Meta's enormous distribution and ecosystem.

I asked Instinct and Muse to compare themselves with each other.

Here are the results.

Instinct output

Muse output

I found Instinct's output more useful.

Caveat: This is one task against a product that is days old. One chart does not crown a winner, it shows where the bar already sits. 

Muse initially generated text and required another prompt to turn the comparison into an image. Sharing it through text also required another step. Instinct produced something closer to what I wanted the first time.

Even the autogenerated filenames were different. Instinct called its file “Instinct vs Muse.png.” Muse called its file “instinct-vs-muse-chart.png.” This is a tiny detail, but tiny details often reveal product philosophy. One feels designed for a person; the other feels designed for a system.

None of these differences is revolutionary. That may be exactly the point.

Good user experience rarely comes from one revolutionary feature. It comes from hundreds of small decisions that remove friction. Those small decisions compound.

Muse has more features and the distribution power of Meta. Instinct is a startup. But, at least for me today, Instinct provides the more natural personal-assistant experience. If my experience with Muse beats my experience with Instinct, the verdict flips. 

But UX is not enough. There is another dimension that may ultimately matter even more than UX: Trust.

The more useful a personal assistant becomes, the more access it needs. Email. Calendar. Contacts. Messages. Passwords. Location. Eventually, money. That creates an interesting trade-off.

The more autonomous we want our assistants to become, the more access and discretion we need to give them. But the more access and discretion we give them, the more we need to trust them.

An assistant that can do everything for you but that you don't trust cannot be very useful. This is also why the question of proactivity matters.

Should my assistant remind me to write a blog post without asking? Sure.

Should it reschedule a meeting? Maybe.

Should it send an email, buy something or move money without asking? Less obvious. 

The best personal assistants will have to learn not only what we want done, but also when they should act and when they should ask.

The boundaries are learnable, and they are not mysterious. Act when the action is reversible and cheap to get wrong: reminders, drafts, holds. Ask when it is not: money, messages sent in your name, anything a human has to undo. And the boundary should move, because trust is earned one good judgment at a time - the same way you hand a new hire bigger and bigger decisions. 

Autonomy is not a setting. It is a balance that grows with evidence. 

A different kind of AI competition

With chatbots, we mostly compared the intelligence of the models. With personal assistants, intelligence is only part of the product.

A good personal assistant needs to understand you, remember what matters to you, know when to act and when to ask, coordinate with other people and systems, earn enough trust to act on your behalf, and then mostly stay out of your way.

That is a very different product problem.

Instinct and Muse are taking different approaches to building a personal AI assistant. ChatGPT, Claude, Gemini, Grok, and others are also evolving from answering questions to taking actions on our behalf.

Given the speed at which this field is moving, something with a better approach could appear next week. That competition will be good for consumers.

The winner in this space will not be the assistant that can do the most. It will be the one you trust enough to stop checking.