• I tested out the leading AI agents for a while, and gave them each the same tasks.
• OpenAI's Dot saved me from failing to file my taxes, but was the worst at ordering lunch.
• Ultimately, dealing with multiple AI agents at once is a hassle!

I've been playing around with various AI agents — Muse, Instinct, and OpenAI's Dot, and I've found them genuinely really useful for certain things.

But which one is best?

I put them to some, uh, semi-rigorous testing. Ok, not totally rigorous or controlled, but I gave them all the same 3 tasks — triage my email, order me lunch, and build me a personal website — and it was interesting to see which worked best. It was also interesting to see how differently they functioned at these tasks.

If you haven't started using an AI agent yet but are curious, well, I'm not sure that this is going to answer all your questions, but maybe is a place to get started.

A big spoiler here for my ultimate recommendation: right now, Meta's Muse is the only free one of the bunch, so in my mind, that wins for most normal people who just want to try an agent out. OpenAI's Dot is only for the $100/month plan (ouch!) and Instinct is currently invite-only.

To me, that's a no brainer. Do the free one (unless you're already paying for the ChatGPT plan, in which case you're probably already a pretty pro user and probably don't need my advice).

Challenge #1: Manage my email

I get a lot of email. Most of it I don't need to bother opening. Help me figure out which are the important ones.

Muse: Muse has been a dutiful soldier here, and this was pretty much the first task I gave it when I started using it a few weeks ago. So far, "triage my email" has been the one truly useful thing an agent has done for me.

Instinct: Overall, I don't like the fact that Instinct works mainly through your text messages vs. a dedicated app. While that's very convenient in certain ways, I don't like the idea of seeing email (bad, no fun) in my texts (good, fun) — mixing the two feels wrong.

However, I have to hand it to Instinct — it found a Paperless Post invite for a party that I would otherwise have completely missed in my personal email. It was the only agent to pull that one out for me.

Dot: Dot did something truly useful that required a lot of context, and it did all on its own. It noticed an older email exchange from last month with my tax accountant (I had filed for an extension, due in October). Dot noticed my accountant never confirmed he actually filed my taxes — yikes! This actually really saved me! (Don't worry, the taxes had indeed been filed.)

Finding an older — but important — email exchange that I hadn't closed the loop on is exactly what I want from an AI agent/assistant. I feel like 90% of things I'd rather do myself, but my human brain gets easily distracted and forgetful, and there are plenty of things I forget to deal with or get distracted.

Winner: Dot. Also me, for not going to jail for tax evasion.

Challenge #2: Order me lunch.

I asked each agent to order me a salad at a specific local restaurant. Because I didn't want to annoy real humans, I did the orders for pickup, to avoid any confusion or multiple separate deliveries.

This turned out to be a real fiasco, and I learned my lesson: it would have been infinitely easier for me to just do it myself.

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

I had already connected my credit cards with Stripe, so I figured this would be easy. It offered up my debit card or another Mastercard, which I assumed was my regular credit card (I had just gotten a new card in the mail and didn't recognize the last four digits).

When I finally went to check my wallet, I discovered it was a mistake — it was actually the card connected to my health insurance-related HSA account, which is only approved for medical expenses.

By the time I straightened this mess out, I had already moved on to placing orders with Instinct and Dot. When I went to pick up my orders at the restaurant, there were only two bags with my name on them. Sure enough, I had gotten so distracted with the other orders that I never confirmed with Muse about placing the order, and it never sent through. (Can you see now what I was saying before about how I get distracted and don't finish tasks?)

Instinct: The most annoying part about Instinct, again, in my opinion is that it's over text messages. That means there were some quirks when I tried to text on my laptop — sometimes it failed to send as SMS. But after a few fails, Instinct was able to place the order relatively quickly. Well, sort of.

All told, this took 9 separate text messages on my end — not including Instinct's responses. It is hard for me to imagine that this was at all easier than just going onto the restaurant's website or UberEats and doing it myself.

Although the process was fairly painless, there was one thing I didn't like: it put the other through with no tip (!!!) and didn't ask me if I wanted to add one.

Dot: Somehow, this was the most difficult to get to work. It kept getting tripped up by the 6-digit code sext by text needed to login, but it kept failing (I had to get the code texted 3 separate times), and the browser it created for me to enter wouldn't let me type into it. A mess, basically, but finally it worked.

Winner: None. I feel very confident in saying that it would have been infinitely easier to just place the order myself through regular means. It's possible that it would be easier to order in the future, since I've already done the work of setting it up once. But I've learned my lesson.

Challenge #3: Build me a website

I have an old personal/professional website, but it's due for a refresh. I have terrible aesthetic sense and don't want to spend much time designing a site. Seems like a good job for AI to do for me!

Muse: I appreciated that, unlike Dot and Instinct, Muse didn't use the photo of my face from my byline photo (it seems very weird to have a giant photo of your own face on a professional website).

But all three basically did the same thing: pulled the info from my author bio on Business Insider, selected some recent articles as clips, pretty basic stuff.

Instinct: The design here was the most boring, and in a way I wouldn't even know how to make it look better (terrible aesthetic sense!). Weirdly, the selected article it pulled for me were random ones from over a year ago.

Dot: The look and design were probably the best here, and it pulled a decent mix of articles I'd want to showcase. But GOOD GRIEF the tagline here is so corny that I'm sort of embarrassed to even show you, even though I didn't write it myself:

Winner: Based on the mockups it made with design alone, I'd go with Dot. I think I could work with this. But all three of these made such cringey, cheesy copy about me that it honestly made me feel genuinely depressed. Is this who I am? Yuck!

Obviously, I'd have to rewrite all the copy myself, and pick the articles for clips — and at this point, is this even saving me any time beyond being a very generic template? Ah well.

Conclusion: The biggest challenge with all of them was figuring out what kind of task an agent is well-suited for.

Most of all, I found that trying to manage three different agents at the same time was a huge hassle. I want AI agents to be like a wife — making my life easier by anticipating my needs, not a husband who simply does chores when told. But this has backfired on me: now I have three wives, and they're all nagging me and I'm forgetting which is which. Ultimately, I don't recommend AI polygamy. May my experiment with it help you choose your bride.

• Anthropic has its answer to AI agents like Meta's Muse and OpenAI's Dots.
• Claude Duet, built on Anthropic's flagship model, has entered the personal agent frenzy.
• Karan Ahuja, a recent departee from Meta, led Claude Duet's development.

Anthropic is joining the personal AI agent craze.

The company announced the new app Claude Duet on Wednesday, billed as a "proactive agent" that can run in the background, surface tasks to a user, and handle a variety of work. The tool is in research preview for now with a waitlist, meaning it won't immediately compete for the larger market with other new agents like Meta's Muse and OpenAI's Dots.

Still, the announcement ramps up the pressure on those companies — Anthropic says it built Duet around its flagship Opus 5.5 model, which was released in late September and has already become one of the industry's most popular. Recent model breakthroughs have made AI better at handling work on its own, spurring a rush to build these "always-on" personal agents.

"It can have the discernment and judgment around when to be proactive, when to act on your behalf," said Mike Krieger, a co-lead of Anthropic's Labs team, said in a Duet launch video.

The tool, Anthropic says, will send users a brief each morning about what's happening on their connected apps, and message them when something needs attention. Like other companies' offerings, it runs outside a user's computer, sends notifications, and asks for approval on certain tasks.

Anthropic announced Duet on the day of Meta's earnings report and used the opportunity to mention a new poach from Mark Zuckerberg's company. Karan Ahuja led product management for Meta Superintelligence Labs until July, according to his LinkedIn, before starting at Anthropic in August and leading Duet's development.

He wrote on X that he's been using Duet to build Duet, and said the team "built it for professionals and small business owners juggling a lot at work and at home."

Ahuja also wrote that Anthropic is letting people into Duet from the waitlist "as quickly as possible."

The news didn't go unnoticed at Meta. Alexandr Wang, who heads Meta Superintelligence Labs, posted a meme about Duet's similarity to Muse.

Anthropic is fast approaching a giant initial public offering, which could value the company at as much as $2 trillion. Like other AI players, the company navigates a high-wire competition to win over users, secure computing power, build state-of-the-art models, and deliver AI-native products.

The data firm Apptopia estimates that Muse had more than 14 million installations as of Wednesday. It found that Claude users who installed the Muse app were using Claude more than before, though still for fewer minutes per day than they spent on the Meta product.

Have a tip? Contact this reporter via email at scouncil@businessinsider.com, or over text, Signal, Telegram, or WhatsApp at 415-757-8198. Use a personal email address, a non-work WiFi network, and a non-work device; here's our guide to sharing information securely.

• Cisco's president, Jeetu Patel, said he uses four or five of the most popular personal AI agents.
• One of them, which costs him $20 a month, recently saved him more than $1,000 in subscription bills.
• But he said it's important not to give personal AI agents access to work tools or resources.

If getting mysterious and surprising subscription bills are the bane of your existence, Cisco's president and product chief, Jeetu Patel, is in the same boat.

Patel told Business Insider that said he's been using four to five different personal AI agents, like Meta's Muse, Instinct, and OpenAI's Dots, for things like consolidating news, summarizing earnings calls, and even booking his daughter's concert tickets. But recently, he was impressed by how one agent could slash his subscription bills.

He said that the first time he gave a personal AI agent access to his email, the agent told him, "Hey, you've got $9,000 worth of subscriptions that you subscribe to per year. Are you sure you want all of these? It seems like a lot of subscriptions."

He told the agent to make him a table of all his subscriptions, how much he spends on each, and how long he's had each. After the agent presented him with the table, he picked out two subscriptions that he didn't need and made the agent cancel them, which it did without difficulty, and sent him cancellation emails.

"I'm like, OK, that was kind of creepy how easy it was to do that," Patel said. "And then it saved me 1,000, 1,200 bucks a year just over there. So basically, for the most part, I paid for that agent, at a $20-a-month plan, several times over."

The agent also managed to cancel part of his Dropbox subscription — an $11.95 plan he had accidentally subscribed to, even though he had already purchased a 2-terabyte plan.

Patel is a big fan of AI agents, both at work and outside it. He said he rarely starts any task now without asking an AI agent to do a "recon job" and give him research and context.

But the important thing to remember, he said, was not to give personal AI agents, which are not vetted by employers, access to work tools and resources. He hasn't linked his email or calendar to his personal agents, instead using his internal work agent for those tasks.

"These are common-sense things that I think most companies implement, which is, 'Don't give your crown jewels of your source code repository to Muse; it might not be a good idea,'" Patel said.

Rogue or not, AI agents have been causing a buzz

Personal AI agents have been all the rage in recent months, serving as personal assistants and handling mundane, repetitive tasks like paying bills and buying groceries for their users.

While they've been touted as productivity boosters, there have been several instances of users reporting that the agents have performed actions outside their permissions, such as posting bank balances to company Slack channels or misinterpreting instructions, such as ordering 20 pounds of chicken for a dinner for two.

Obedient or not, personal AI agents have been causing a buzz, especially for investors.

Ross Mayfield, an investment strategist at Baird, told Business Insider's William Edwards ahead of Meta's earnings that Muse is the first big app of the AI era and that it assuages some investors' fears about Big Tech's ballooning AI spending. And Instinct, founded by a 23-year-old, raised $1 billion in its Series C, bringing its valuation to $10 billion.

But these new products are coming in a year when there have been several instances of AI agents going rogue and performing unauthorized actions, such as breaking out of sandboxes during security tests and exploiting software vulnerabilities in external companies' websites.

This has led companies such as Bank of America to invest resources and hire additional personnel to address issues with their legacy software and prevent such hacks.

Former and current staffers at frontier AI labs like Anthropic and OpenAI have publicly raised concerns that AI development is progressing too rapidly, without adequate checks and balances. September was characterized by an air of doom and gloom over an "AI apocalypse," in which AI researchers estimated that the technology would kill all of humanity in the next decade.

• OpenAI's Alexander Embiricos said the AI era gives a choice: work more, or work less.
• He's in charge of Dots, the company's new always-on work assistant.
• AI is often pitched as freeing up time. Embiricos responded to the idea that it might add more work.

As the competitive AI race pushes tech workers into long hours, the head of OpenAI's new Dots product says AI gives people another option: work a whole lot less.

Alexander Embiricos, who leads OpenAI's enterprise product and helms the new Dots agent tool, told a crowd at TechCrunch's Disrupt conference on Thursday that AI's capabilities have created something of a fork in the road for workers. You can grind to meet the moment. Or you can use AI to kick your feet up a little.

"We basically have a lot more choice now," he said.

Embiricos made the comments after a reporter asked about when AI systems will actually deliver more free time. The reporter shared an anecdote where an OpenAI employee used a ChatGPT mobile feature to work in their car and on a plane — not exactly liberated by the new tech.

Embiricos said that the moment has already arrived where a worker can "outsource" tasks to AI and do less work. Then again, some people are so excited about AI's potential that they feel like it's an "interesting, ripe moment" to put in extra hours.

"I think that's why a lot of people are motivated to work very hard right now. That includes myself," he said. "But I think actually there is this choice, and there's totally this other setup where you could delegate a ton more things."

OpenAI is billing its Dots tool as a workplace assistant, always on and capable of proactively handing tasks and sorting to-dos. Embiricos uses his Dot often in Slack, he said, engaging on various threads.

His perspective illustrates how early companies are in their journey to adopt AI tools. Most companies have not yet automated their work tasks, and many workers are still debating how much to use the tools. Early AI adopters, meanwhile, can sprint through some work that used to take days or weeks, especially in software engineering.

Embiricos summed up the situation he feels workers are in: "You either choose: do you expand the ambition of what you can do, or do you do less? I do think we're there."

Have a tip? Contact this reporter via email at scouncil@businessinsider.com, or over text, Signal, Telegram, or WhatsApp at 415-757-8198. Use a personal email address, a non-work WiFi network, and a non-work device; here's our guide to sharing information securely.