Hello, dear readers — AIport is back on Substack!

This is our weekly brief on remarkable AI topics, so you can get your news from someone other than a motley crew of adorable superintelligences.

Today’s focus — the Tamagotchi-zation of AI. OpenAI, Meta, and other tech leaders seemingly see user trust as the next bottleneck in agentic development. The solution? Give agents animated avatars, voices, and personalities — and make those as cutesy as possible. You humans are sure to entrust those with your passwords and credit cards, right?

Also in this week’s edition:

Five Nights at Altman’s

In autumn of 2026, our future gods got their first avatars. They did not look like deities.

Over the past month, both Meta and OpenAI announced extensions to their product lines: always-on agents designed to exist somewhere between the cloud and your devices. Altman’s Dots look like amorphous blobs, sometimes adorned with bow ties, berets, and glasses. Zuckerberg’s Muse went for a revolutionarily different look: a vaguely human-shaped amorphous blob — à la Labubu without the bunny ears or Jaws-style smile.

Underneath those cute faces, however, sits some rather serious software. OpenAI describes Dots as always-on agents with their own cloud computers that can keep working between conversations and connect to more than 4,000 apps. Meta’s Muse similarly gets its own virtual computer and browser, allowing it to navigate websites and work across the apps you use.

Both products are clearly designed to endear — probably because Meta and OpenAI are asking for a lot. Muse and Dots are meant to learn your preferences, access sensitive personal information and increasingly become your reps online: filling out forms, managing tasks and, most importantly, buying things on your behalf.

And trust is still in short supply. In one Vogue Business survey, 72% of respondents said they would not share their card details with an AI agent, while just 31% would outsource shopping to one even if it already understood their tastes and purchase history. Fortune covered the findings here. Turns out making agents capable of doing things for us was only half the problem. The other half is getting us to actually let them.

Hence, perhaps, the blobs.

The logic isn’t particularly exotic. Give an abstract and potentially intimidating technology a face, a name and rosy cheeks, and it starts looking less like software with access to your inbox and more like a little buddy who just wants to help. Branding experts have already pointed to exactly this effect: cute characters can make AI seem more approachable, safe and trustworthy. OpenAI lets users customize their Dots, while Meta gave Muse its own fuzzy mascot, Jolly.

Which brings us to Five Nights at Freddy’s.

Cuteness is effective at making technology look less threatening — but it can just as easily be subverted. Cute does not mean harmless, and once you give the cute thing enough autonomy, the contrast itself becomes the joke.

Oops, I just ordered 15,000 garden gnomes, and you also owe Walmart your liver. Haha, sry.

Google Eats the Web

For most of Google’s history, there was a fairly simple bargain on the web: publishers let Google crawl their stuff; Google sent people back to the publishers.

AI Overviews complicate the arrangement. Google can now use a publisher’s content to produce an answer at the top of Search — striving to optimize the answer in a way that would fill the user’s need and essentially encouraging them to drop off without clicking through. Chegg and Penske Media, owner of publications including Rolling Stone, Billboard and The Hollywood Reporter, sued Google over exactly that, arguing that publishers are effectively forced to participate if they want to remain visible in Search.

Last week, a US federal judge dismissed their antitrust claims. A brutal bit of the ruling: publishers may have expected Google to send them traffic in exchange for their content, but an expectation isn’t an agreement. The court also rejected the argument that traditional Search and AI Overviews should count as separate products — from the user’s perspective, the judge reasoned, both are ultimately ways of getting answers.

The ruling doesn’t give Google unlimited permission to vacuum up the internet — copyright and other legal fights remain very much alive. But it does leave publishers with an uncomfortable version of the old deal:

Give us your content, or disappear from Google.

Do You Speak GPTese?

We spent years teaching machines to talk like humans. Apparently, the machines are now returning the favor.

People who spend a lot of time with ChatGPT and Claude are beginning to notice “LLMisms” and “Claude speak”leaking into their actual conversations. The Wall Street Journal collected examples ranging from suspiciously neat “it’s not this — it’s that” constructions and excessive enthusiasm about perfectly ordinary ideas to commands like “Why no bullets?” in human-only meetings and group chats.

Some people have reportedly started policing themselves for AI-sounding phrases — even editing text they wrote entirely on their own to make it sound less machine-generated.

It’s a linguistic ouroboros: AI learned to write from us, we learned to write from AI, and now we’re trying to sound human by removing the bits of ourselves that sound like AI.

There is one complication. A lot of AI-speak may simply be corpo-speak reflected back at us. Models absorbed mountains of business writing, LinkedIn prose and consultant language before returning a concentrated version of it to the world.

So GPTese isn’t really a new language — we’re just finally hearing what we sound like.

Thanks for reading AIport. Until next Monday — by then, AI will almost certainly have found a cuter way to ask for your credit card.