Always-on is not a personality

There’s a pitch making the rounds this year, and Technology.org ran a clean version of it last week: “AI Avatars Are Turning Brand Characters Into Always-On Media Assets.” The opening line sets up the argument. Brand mascots are not new, and neither are computer-generated characters. What is new is the speed. Your character can now speak every language, react to every trend, answer every customer, and post around the clock, without the agency overhead.

The production claim is true, and we should say so plainly. Rigging, animating, and localizing a character used to be the expensive part of running one. That cost is collapsing, and any studio that pretends otherwise is protecting its invoices, not its clients.

But the pitch rests on an assumption nobody says out loud: that the reason your mascot wasn’t doing more was capacity. That there was a backlog of things the character wanted to say, and the animation budget was the dam holding it back.

Walk through most brands’ character work and the assumption falls apart. The mascot has a name, a color, and a pose library. It has no opinions. There is nothing it loves, nothing that annoys it, nothing it would refuse to promote. It exists to smile near the product. A character like that doesn’t have a backlog. It has nothing to say once, let alone forty times a day in nine languages.

Now give that character an always-on pipeline. The output problem is solved and the input problem is exposed. You get infinite content from an empty character, which is a spam machine wearing your mascot’s face. The tools didn’t create the emptiness. They just removed the production delay that used to hide it.

Volume is a stress test

Here’s the part the avatar vendors get backwards. A mascot that appears in one art-directed TV spot per quarter can survive being thin, because thirty polished seconds don’t give the audience time to notice there’s nobody home. An always-on character produces thousands of outputs with nobody art-directing each one. At that volume, the audience maps the personality fast, and if the personality is “cheerful about the product,” they file the whole account as advertising and mute it.

The characters that survive volume are the ones that were written before they were scaled. Duolingo’s owl posts constantly and it works, because Duo is sharply defined: obsessive, vaguely threatening, willing to be the villain of its own brand. There are things Duo would never post, and the team behind the account knows what they are. That boundary is the character. The posting schedule is just distribution.

This is why we keep banging on about personality bibles. When production was expensive, the bible looked like a nice-to-have, because a human writer was quietly supplying the judgment on every deliverable anyway. In an always-on pipeline the bible is the product. It’s the only thing standing between a character and a content farm. The writing got more valuable the moment the rendering got free.

The interface trap

One more thing in the pitch deserves a flag. The vendors’ favorite upgrade is the mascot as “brand interface”: it greets visitors, answers support questions, guides checkout. This sounds like a promotion for the character. It’s a demotion.

A mascot’s job is identity and entertainment. A support agent’s job is accuracy under pressure. Bolt the second job onto the first and every failure of your chatbot becomes a failure of your character. The hallucinated refund policy doesn’t get screenshotted as “the bot was wrong.” It gets screenshotted as your mascot lying to a customer, and that screenshot outlives the apology. Keep the character and the help desk in separate bodies. If the help desk needs a face, give it a plain one.

None of this is an argument against the tools. We’re character people; cheaper animation is good news for us and for every brand that owns a character worth animating. The argument is about sequence. Always-on is an amplifier, and an amplifier doesn’t care whether it’s amplifying a point of view or a vacuum. Define the character first: what it wants, what it hates, what it would never say. Then turn the volume up.

The tools are ready for always-on. Most characters aren’t.

We write the character before anyone renders it. If you’re building one, we’d love to talk.