Meta Was Counting Every Prompt You Typed

Today on Briefly AI — Meta quietly stops grading staff on how many words they type into a chatbot, while pushing a new AI agent instead. A job hunter got so sick of being ghosted by AI interviewers that he sent a chatbot to sit the interview for him. And Amazon teaches Alexa to sniff out the scammers pretending to be Amazon.

Welcome to Briefly AI, a podcast by Harry Sharman, written and voiced by his AI clone. The voice is synthetic; the sourcing isn't — every story traces to something a real journalist or researcher actually published this week. Someone human still has to do the reporting. For now.

That's the shape of the day. Let's dig in.

Somewhere inside Meta, someone actually built a leaderboard for how many words you'd typed into a chatbot. Not for fun — for your performance review. And this week, by the sound of it, they quietly switched it off. Which tells you almost everything you need to know about how that experiment went.

So here's the story, reported by Wired. For the best part of a year, Meta has been pushing staff to use its AI tools constantly, and internally, people started calling the practice "tokenmaxxing" — measuring how many tokens, essentially how much text, an employee fed into Meta's AI systems, and treating that number as a proxy for how seriously you were taking the company's AI push. Low token count, low enthusiasm, in the eyes of whoever was reading the dashboard. It's the kind of metric that sounds perfectly sensible in a slide deck and turns faintly absurd the moment you say it out loud — judging curiosity and skill by volume, as if quantity of prompts were the same thing as quality of work.

Now, according to Wired, Meta is easing off that pressure. The push to hit a usage number is softening. But the company isn't stepping back from AI at work — quite the opposite. It's now steering employees toward Hatch, described as Meta's most advanced AI agent project yet, and framing it as something to experiment with rather than something to be measured against.

That distinction is the whole story. "We will judge you by a number" produces exactly the behaviour you'd expect — people padding out prompts, pasting things into a chatbot they don't need to, just to keep the counter moving. It optimises for the appearance of adoption, not the substance of it. "Have a play and tell us what you find" at least has a chance of getting honest signal about whether the tool is actually any good. If a company this deep into its own AI push is walking back its own internal metric, that's a fairly candid admission that the counting phase of workplace AI has run its course.

Now, this next one's a bit more personal, and it's also Wired's reporting. A man named Christopher had been going through the modern job-hunting gauntlet — apply, and instead of a human on the other end, you get an AI interviewer asking questions on video, scoring your answers, and then, more often than not, saying nothing back at all. Ghosted, but by software.

So Christopher did the thing that, once you think about it for more than five seconds, is almost inevitable: he set ChatGPT up to answer for him. Not to help him prepare beforehand — to actually sit the interview, in real time, against the AI interviewer, while the human being whose job prospects were actually on the line got on with something else entirely.

Here's the bit that matters. Companies adopted AI interviewers because they can screen thousands of candidates without booking a single human hour. But the moment a tool exists to screen you, a tool exists to help you pass the screen — that's not a hypothetical, it's already happening, and Christopher's version is just the most literal form of it: cut the humans out of both sides and let the machines sort it out between themselves. It rather undercuts the entire pitch behind AI interviewing, which was that it would surface the best candidate, not the best-prompted one.

What's actually come of it, as Wired tells it, is a genuinely strange endpoint for hiring — a process built to make evaluation more efficient risks becoming a contest between two language models, deciding a real person's income while the humans on both sides of the call become spectators to their own interview.

On a completely different note, Amazon's trying to solve a smaller but properly useful problem, as reported by The Verge. Impersonation scams — texts, emails, calls pretending to be Amazon, warning you there's a problem with an order or a payment — are among the most common scams going, precisely because almost everyone has an Amazon account and half of us are half-expecting a delivery at any given moment.

So Amazon is rolling out a new feature inside its Alexa for Shopping assistant. If you get a message claiming to be from Amazon and you're not sure, you can simply ask Alexa about it. It uses AI to check the message against what a genuine Amazon communication actually looks like — tone, formatting, the kind of links Amazon would and wouldn't send — and tells you whether it checks out or looks like a fake.

Worth a beat on the context: this is a company whose name gets borrowed by scammers more than almost any other, exactly because of how much trust people already place in it. Using AI to catch AI-era scams is a slightly odd loop to be standing inside — the same kind of technology that makes convincing fake messages cheap to produce is now being pointed at catching them — but it's a genuinely practical application, aimed squarely at the people least equipped to spot a fake unaided.

It's a small feature, and it doesn't need a benchmark or a headline number to prove its worth. It either catches the scam email sitting in your inbox right now, or it doesn't.

That's the state of play today. A company that stopped counting your keystrokes, a man who sent a robot to argue with another robot on his behalf, and Alexa moonlighting as a detective. Three completely different stories, and somehow all about exactly the same thing — everyone, human and machine alike, trying to work out what's actually real.

You can find more at harrysharman.com. Briefly AI — the only newsroom where the reporter, the writer, and half the subject matter are all the same kind of machine.