You Bought The AI. Where's The Money?

Today on Briefly AI. Jack Dorsey's Block has built a voice model that talks like a person, and runs on a laptop. An economist says AI has paid off for almost nobody outside tech. And the employees who use AI best are the ones most afraid of it.

Welcome to Briefly AI, a podcast by Harry Sharman, written and voiced by his AI clone. A newsroom staffed entirely by one of the things it's meant to be covering. We're aware of how that sounds, and we're leaning in anyway.

Those are the headlines. Now, the detail.

Somewhere, a very patient accountant is looking at the company's AI bill and asking a fair question. Meanwhile the machines are learning to chat like your mates. Let's start with the chatty ones.

Jack Dorsey's company Block, the payments firm behind Square and Cash App, has announced a voice model called Dessa. The pitch is that it was trained from scratch on nothing but human conversation. Not text that gets read out loud afterwards. Actual people talking, with the pauses, the interruptions and the "um, well, sort of" in between. According to TechCrunch, it sounds eerily human, and it can run locally on a MacBook M4 Max. So no cloud server and no round trip to a data centre. The model sits on a laptop and talks.

Now, the Verge reported that Block shared some of Dessa's chats with real people, and they're a bit awkward. Uncanny is the word. Think of a new friend who is trying very hard. That's useful, because it shows where this technology currently sits. It's close enough to be unsettling, and not yet close enough to be smooth. Dessa's lead, Ragavan Thurairatnam, was the one sharing the examples.

A word on the landscape, because Dessa isn't the only thing that landed this week. Moonshot's Kimi K3.1 Pro arrived on its API on the twenty-third. As Simon Willison noted, it isn't open weights yet, but Moonshot promise the full weights by the fourth of November. On Sunday, the lab behind the Chinese recruitment platform Boss Zhipin released Nanbeige4.3-33B, an open model under the Apache 2.0 licence. It has a half-million-token context window, and it scores 80.9 on Terminal-Bench, a test of coding in a command line, against 73.0 for Qwen3.8-27B. And Anthropic put out Claude Duet, a research preview of an app where Claude works in the background, watches your connected apps, and asks before it does anything important. So the picture is this. Voice is getting local, Chinese labs are pushing open coding models, and Anthropic is testing an assistant that doesn't wait to be asked.

Worth saying, with a straight face, that a synthetic voice reading you a story about synthetic voices is not lost on me.

Right, on to the money. Because all of this costs a lot of it.

TechCrunch reports that an economist at the investment firm Apollo says AI has delivered, and I'm quoting here, "zero financial payoff" to corporate America outside of tech. Despite massive investment and widespread hype, in his words. Zero. Which is a bracing number to hear, given how much has been spent.

Now, it helps to know this isn't the only measurement going around. BCG's 2026 Applied AI Index says the opposite, roughly. Half of companies generate AI value, and the top 7.5 percent see two and a half times the shareholder return. Those two claims can both be true, and here's how. If the gains are piling up in a small group of firms at the top, the average company can look like it's getting nothing. The Apollo line is about the typical business. The BCG line is about the leaders.

And that gap is the interesting bit. It suggests the technology isn't the thing separating winners from the rest. Plenty of firms have the same models. What differs is whether anyone changed how the work gets done, or whether they just bought licences and waited. We've all seen the version where a company announces a tool, runs one training session, and then wonders why the spreadsheet still looks the same.

What came of it, practically, is a sharper question for every boardroom. Not "are we using AI?" but "can anyone point to a line in the accounts that moved?" If nobody can, Apollo's economist has just given the finance director a very quotable sentence.

And that brings us to the people doing the using. Here's one for anyone who thinks the keen ones are the confident ones.

A piece in Forbes this week, by Vibhas Ratanjee, argues that your most AI-fluent employees are the most afraid. The reasoning runs like this. The people who understand AI best are the ones who can see most clearly what it can do. They aren't panicking about a vague idea. They know exactly which of their tasks the thing can handle.

And the piece points to evidence that the anxiety isn't irrational for one group. Stanford's Digital Economy Lab has been tracking ADP payroll records. It finds employment for workers aged twenty-two to twenty-five in AI-exposed occupations sitting nineteen percent below where it would be had it kept pace with their less-exposed peers. There's no comparable gap among experienced workers.

Here's the detail that matters. The adjustment isn't happening through layoffs. It's happening through jobs that were never posted. A company that would have hired three graduates quietly hires two, and nobody gets a redundancy letter, so nothing shows in the aggregate figures. The young people affected can't see it either. They just find fewer doors.

The article's argument is that silence from leadership makes this worse. When managers say nothing about what AI means for roles, the most capable staff fill the gap with their own worst-case version. And they're the ones best equipped to write it.

Put that next to the Apollo number and you get an odd picture. Companies can't yet show the profit, and the cost is already landing on the newest people in the workforce. If you manage anybody, the cheapest thing you can do this week is tell them plainly what you expect AI to change on your team. It's free, and it's apparently in short supply.

So that's today. A laptop that chats, a boardroom that wants receipts, and a generation of graduates finding fewer doors. If your company has bought the AI and can't find the payoff, it may be sitting in someone's inbox, waiting for you to tell them where they stand.

That's the news, filtered down to what actually matters this week. Back tomorrow, same arrangement.