Why AI agents became cute

Read time: 5 mins 2 secs

Presented by beehiiv

You already know I’ve used beehiiv to grow Brand Chemistry from zero subscribers to 120,000 today. Connecting the MCP to Claude has completely changed how I operate.

Here’s three prompts I used this week:

None of these are clever questions. I just didn’t have time to dig for the answers before. Now, I get the answers and even actions done for me in a few minutes.

Not the first time beehiiv have shipped something that makes my work way more efficient, and based on their recent updates it definitely won’t be the last either.

Get 30% off your first 3 months with code ISAAC30.

Thank you for supporting our sponsors, who keep this newsletter free.

Interested in sponsoring these emails? See our partnership options here.

A few weeks ago Mark Zuckerberg walked on stage and introduced Jolly, a fluffy character with beady black eyes and a little upturned smile. It's the face of Muse, the personal AI agent Meta launched two weeks earlier. The same day, Meta's Chief AI Officer Alexandr Wang was posting Jolly memes on X.

Jolly had plenty of company. Less than a week later, OpenAI launched Dots, where you can name your agent and dress its character in accessories like a beret or a bow tie. Only a month earlier, xAI (Elon Musk's AI company) launched their GrokBot agents which let you pick a shape and colour for each agent's character. Three of the biggest AI companies launched always-on personal agents within a few weeks of each other, and all three put a soft, simple character on the front.

Brand characters are one of the classic advertising moves, but it’s interesting to explore why personal AI agents have all flocked to run the playbook.

Ad-testing firm System1 went through every long-term campaign entered into the IPA Effectiveness Awards, the UK's main advertising effectiveness awards, since 1998. Of the campaigns built around a character, 41% reported very large gains in market share, against 30% of the ones without. For profit it was 34% against 26%.

Characters were reliably driving improved performance for brands, but they still fell out of fashion. Among long-term campaigns entered into the same awards, the share built on a character or a slogan fell from 41% in 1992 to 12% by the mid-2010s. For a while, mascots were only really used by cereal brands.

Duolingo has been a standout exception, but even then, they announced they were toning down their unhinged owl on social this year. I called that a big mistake, because people followed the chaotically entertaining character far more than they followed the language app. A character is powerful for giving people something to recognise and something to emotionally connect with, especially in product categories that give them very little of either.

That’s the reason another category that never let go of characters is insurance. You can't see the product, since you're buying a promise you hope you never have to test. You can't tell one policy from the next, because they're written in the same fine-print legalese nobody reads. You have to trust the company to help you on what might be one of the worst days of your life. So GEICO, the US car insurer, have a gecko, and Compare the Market, the comparison site, have a meerkat.

A personal AI agent faces all three of the same problems as insurance brands, arguably to an even larger extent.

Grok Bot, Muse and dots all do their work on a cloud computer you don’t look at. You hand over a job and get back a result, and most of the work happens somewhere you never actually see.

A character gives that work a tangible form. X's design team built Grok Bot's icon in code, so its waiting and working states fluidly morph into each other, and you can give each agent its own shape and colour so you can see at a glance which one is busy. OpenAI is already describing teams of dots, which be the same thing. Meta has gone a step further and made a Tamagotchi-like wearable for Muse, so the agent has a physical embodiment outside your phone or laptop.

The character becomes the embodiment of the agent, it’s the tangible element you can actually see and interact with.

On paper, the three personal agents are almost identical. They run around the clock, they sign into your accounts and finish multi-step jobs while you do something else. When the products are completely commoditised, a character is just about the fastest way for people to recognise something as yours.

British Gas, the UK energy supplier, ran the same "Free Electricity" ad with and without its penguin mascot, Wilbur, and measured the results. With the character in the ad, spontaneous brand recall rose from 28% to 35%.

Meta have dialled up Jolly’s social presence to show exactly how far that impact goes. People have made it their own, sharing their own pics of Jolly, including some pretty NSFW ones. Either way, Jolly is showing up all over social feeds and connecting people back to Muse, which a logo struggles to do. When GitHub and Spotify announced their Muse integrations, they posted their own versions of Jolly.

The features of these personal agents will keep converging. Jolly, and all the emotional attachment people develop with them (him? her? it?), will still only belong to Meta.

Muse connects to your email and calendar, and it can even check out using your credit card. That's a much bigger personal data ask than an insurance policy requires, and it's the problem these characters work hardest on.

The first way they do it is by disarming you. In 2009, a study published in the journal Ethology showed people photos of infant faces with "baby schema" features dialled up and down, meaning the round head, big eyes and small mouth. They discovered that the more of those baby features a face had, the cuter people thought it was and the more they wanted to look after it. Jolly, dots and Grok Bot are all round and very simple, closer to a toy than a person. They utilise our psychological pre-disposition so the thing accessing your bank account looks as cute and trustworthy as possible, like something you'd actually want to look after and care for.

The second way they do it is by making it your own. People value things more when they've had a hand in making them. Researchers call it the IKEA effect, after a 2012 study where people valued storage boxes they'd assembled themselves more highly than identical ones built for them. Notion did this years ago when they let people personalise their Notion AI agent with accessories and even name them. The more effort you put into customising how the agent looks, the more it feels like yours to keep, and the more you value it.

There’s been significant pushback to these cute and cuddly characters. Ed Zitron, a PR founder and vocal AI critic, told Fortune the mascots are "entirely marketing". Fairplay, a children's advocacy group, accused Meta and OpenAI of using "children's iconography" to sell AI products.

Part of the concern is that the characters drastically oversimplify and at times even misconstrue the actual technology. For example, Grok Bot lets you give every agent its own name, shape and colour, which makes each one feel like a separate little being. But, their documentation says all of your agents use the same cloud computer, "sharing its files, browser sessions, and app logins".

Muse especially has exploded in popularity, passing 5 million downloads in its first three weeks. However, Muse’s “Confidential VM”, which is the part that would lock your data behind a key only you hold so that even Meta couldn't read it, hasn’t been launched yet. It’s due "later this year".

So these characters are building trust and asking for access to private information before the technology has proven it can actually handle it. That trust is a loan, and it's the product that has to pay it back by following through.

If people can't see your product working, can't tell it apart from a competitor, or have to trust it with something that matters, then designing characters can be a useful way to address that problem. Just make sure the product can follow through on what the character promises. Meta can put Jolly on stage this month, but its Confidential VM is the part people will judge it on longer term.

The characters make these personal agents easy to like. Whether people keep handing over their private information depends on what ships underneath them, and without that information access these AI products won’t work. Brand characters are once again a critical and highly-valuable asset for the companies behind them.

If you enjoyed this post or know someone who may find it useful, please share it with them and encourage them to subscribe: brandchemistry.co/p/why-ai-agents-became-cute