I tested Muse, Dot and Instinct. The outcome is clear.

I’ve been trying out various AI agents – Muse, Instinct, and OpenAI’s Dot – and I’ve found them to be really useful for some things.

But which one is best?

I put them through some, uh, semi-rigorous testing. Well, not exactly strict or controlled, but I gave them the same 3 tasks – sorting my emails, ordering me lunch, building me a personal website – and it was fun to see which one worked best. It will also be interesting to see how they perform differently on these tasks.

If you haven’t started using an AI agent yet but are curious, well, I’m not sure this answers all of your questions, but maybe it’s a place to start.

There’s a big spoiler here for my final recommendation: Meta’s Muse is currently the only free one of the bunch, so in my opinion, that’s a win for most regular people who just want to try proxies. OpenAI’s Dot is only available on a $100/month plan (yikes!), while Instinct is currently invite-only.

To me, this is a given. Do it for free (unless you’ve paid for the ChatGPT plan, in which case you’re probably already a very expert user and probably don’t need my advice).

Challenge #1: Manage my email

I get a lot of emails. Most of them I didn’t need to bother opening. Help me figure out which ones are important.

meditation: Muse has been a dedicated soldier here, and this was pretty much the first mission I gave it when I started using it a few weeks ago. “Categorize my emails” is the one really useful thing the agent has done for me so far.

instinct: Overall, I don’t like that Instinct works primarily through text messages rather than a dedicated app. While this is very convenient in some ways, I don’t like seeing emails (bad, boring) mixed in with my text (good, interesting) – it just doesn’t feel right to mix the two.

However, I have to leave it to instinct to spot a paperless party invitation that otherwise I might have missed entirely in my personal email. This was the only agent who pulled that out for me.

point: Dot does some really useful stuff that requires a lot of background information, and it does it all on its own. It noted an older email exchange with my tax accountant from last month (I had applied for an extension, which was due in October). Dot noticed that my accountant never confirmed that he actually filed my taxes – oops! This really, really saved me! (Don’t worry, the taxes have indeed been filed.)

Finding an older but important email exchange that I haven’t closed the loop on is exactly what I want from an AI agent/assistant. I feel like 90% of things I would rather do myself, but my human brain is easily distracted and forgetful, and there are a lot of things I forget to process or get distracted by.

Winner: Points. And me, for not going to jail for tax evasion.

Challenge #2: Order me lunch.

I asked each agent to order a salad for me at a specific local restaurant. Since I didn’t want to annoy a real person, I placed a pickup order to avoid any confusion or multiple separate deliveries.

It turned out to be a real fiasco and I learned my lesson: it would be much easier if I did it myself.

meditation: Initially, Muse was very helpful. I accidentally gave the wrong restaurant name, but Muse helped me find the right place and pulled up the correct menu options for me to review.

I already connected my credit card to Stripe, so I thought it would be easy. It provides either my debit card or another MasterCard, which I assume is my regular credit card (I just received a new card in the mail and don’t recognize the last four digits).

When I last checked my wallet, I discovered it was a mistake—it was actually the card associated with my health insurance-linked HSA account, which was only approved for medical expenses.

While I was sorting this mess out, I was already placing orders with Instinct and Dot. When I went to the restaurant to pick up my order, there were only two bags with my name on them. Sure enough, I had been so distracted by other orders that I never confirmed to Muse that I had placed the order, and it was never shipped. (Can you see now what I said earlier about how I get distracted and fail to complete tasks?)

instinct: In my opinion, the most annoying part of Basic Instinct is via text messages. This means something weird happens when I try to send a text message on my laptop – sometimes it doesn’t work as a text message. But after a few false starts, Instinct was able to place orders relatively quickly. Well, sort of.

All told, I received a total of 9 separate text messages – not including Instinct’s reply. I can’t imagine this being easier than logging onto the restaurant’s website or UberEats and doing it yourself.

While the process was pretty painless, there was one thing I didn’t like: it left another thing without a prompt (!!!) and without asking me if I wanted to add one.

point: Somehow this is the hardest to come to work. It keeps getting stuck with the 6-digit code sext required to log in, but it keeps failing (I have to send the code 3 separate times), and the browser it creates for me won’t let me type it. It was basically a mess, but it worked out in the end.

Winner: None. I feel very confident in saying that it would be easier to place the order yourself through the regular means. It might be easier to order in the future since I’ve already done the setup once. But I’ve learned my lesson.

Challenge #3: Build a website for me

I have an old personal/professional website that needs a refresh. I have poor aesthetics and don’t want to spend too much time designing a website. Looks like artificial intelligence can do a great job for me!

meditation: What I appreciate is that, unlike “Dot” and “Instinct,” “Muse” doesn’t use a photo of my face in my byline photo (it seems odd to have a giant photo of my own face on a professional website).

But all three basically did the same thing: pulled information from my author profile on Business Insider, selected some recent articles to clip, pretty basic stuff.

instinct: The design here is the most boring, to the point where I don’t even know how to make it look better (terrible aesthetics!). Oddly, the article it picked up for me was a random article from over a year ago.

point: The look and design is probably the best here and it contains a good mix of articles that I want to showcase. But gosh, the slogan here is so corny I’m embarrassed to even show it to you, even though I didn’t write it myself:

winner: Based on the model being made with just the design, I would choose the Dot. I think I could use this. But all three struck me as awkward, cheesy, and honestly, really frustrating to me. Is this me? oops!

Obviously I’d have to rewrite all the copy myself, and choose to clip the articles – at this point, besides being a very versatile template, would this save me time? Ah okay.

in conclusion: Their biggest challenge is figuring out what tasks agents are suited to.

On top of that, I found it to be a huge hassle trying to manage three different agents at the same time. I want an AI agent to be like a wife—making my life easier by anticipating my needs—rather than a husband who just does what I’m told to do around the house. But it backfired: now I have three wives, and they all nag me so hard that I forget which one is which. Finally, I do not recommend AI polygamy. I hope my experiment helps you choose your bride.