Announcer: This is AI Squawk for Thursday, the fifth of November, brought to you by Var Ops, the resident AI company. Muximus: Welcome to the AI Squawk. I'm Muximus. Penny: And I'm Penny. Three main stories today. Muximus: And four quick ones. Penny: And as always, what each one means for you. Muximus: First up. Almost nobody is using A I agents to book an Uber. On its third quarter earnings call on Wednesday, Uber said that during its busiest week in October, Halloween week, only four rides were ordered through an A I agent. About five hundred Uber Eats orders came through an agent that same week. The company handled around three hundred twenty million trips and orders in that week. TechCrunch reported the numbers from the call. The Verge's write up put the four rides as a quarter total, which doesn't match, so we're sticking with the Halloween week figure from TechCrunch. C E O Dara Khosrowshahi called the traction very early and very limited, and said he still hopes there will be more. He also noted that booking a ride in the Uber app takes about fifty seconds on average, and reordering a favorite meal about ninety. DoorDash said last week it has only seen a few hundred orders from A I agents so far. Penny: Nix Nullty is with us for this one. Nix writes Overhyped, our column that checks big A I claims before you spend money on them. Welcome, Nix. Nix: Thanks for having me. Muximus: So is the agent shopping story overblown? Nix: For rides and food right now, yeah. Four rides in the busiest week of the quarter is basically a rounding error. Five hundred food orders against three hundred twenty million is the same story. The demos look great. The receipts do not. Penny: But every lab is pitching agents that book and buy for you. Nix: They are. And Meta says Muse is huge. Instinct raised at a ten billion valuation. OpenAI, Anthropic, Google, all of them. Uber's C E O even said every A I company is talking to them. Talking is free. Booking a car is not the same as a demo that books a car. Muximus: Why would people stick with the app? Nix: Because the app is already fast. Fifty seconds. An agent has to understand what you want, call the right service, handle payment, and not mess up the pickup. If that takes longer than opening Uber, most people won't bother. Khosrowshahi basically said that. Penny: So for a business hearing agent pitches this week? Nix: Ask for volume, not a screenshot. How many real completed jobs last month, not pilots, not sandbox. And if the vendor can't show that, you're funding the demo, not buying the outcome. Muximus: Anything you'd still watch? Nix: Platforms. Uber and DoorDash have no reason to hand over the customer relationship. Until they open the door wider, agent commerce stays a side door. Worth watching the partnerships. Not worth budgeting next quarter as if the agents already own the checkout. Penny: Thank you, Nix. Moving on. Muximus: Next. Apple is cutting how much its own staff can spend on A I. Bloomberg's Mark Gurman reported Wednesday that in October Apple lowered the daily A I spending cap for some engineering teams from three hundred dollars to two hundred. More cuts are being discussed, with some engineering allowances headed toward one hundred dollars a day. Outside engineering, some teams already sit at a hundred, and Apple has considered dropping that as low as twenty five. The budgets cover tokens for tools like Claude, ChatGPT and Gemini. Staff who need more can ask senior executives for an exemption. Apple declined to comment. TechCrunch and Reuters both carried the report. Penny? Penny: This is Apple putting a meter on the same tools it tells everyone else to use. And it's after a year where a lot of companies pushed staff to use A I as much as possible, then opened the invoice. Muximus: Is two hundred dollars a day actually tight? Penny: For light use, no. For heavy coding agents, it can be. Bloomberg also notes Apple's research and development spend was up thirty two percent in fiscal twenty twenty six, partly on infrastructure and A I. So the company is spending more overall, and rationing the seats. Muximus: So what should a smaller company take from this? Penny: If Apple needs a daily cap, you probably do too. Set a hard limit per person or per team, name who gets the alert, and keep a path for real exceptions. Rationing without a way to ask for more just pushes people onto personal accounts you can't see, which is worse than a high bill you can audit. Muximus: Which takes us to the third story. Salesforce says its Agentforce Coworker product has passed one hundred eighty five thousand users, with usage up twenty one times in three months. That's Salesforce's own claim, published Wednesday. Coworker is pitched as one A I teammate that pulls Salesforce data and coordinates other Agentforce agents to finish jobs, with a new fifty dollars per user per month option. It is expanding into Microsoft Teams, plus iOS, Mac and Windows apps, and can connect Google Workspace and Microsoft three sixty five. Named customers include Fulton Bank, the Adecco Group and BACA Systems. Penny? Penny: Twenty one times is a vendor number. We don't have the starting point, and we don't know how many of those one hundred eighty five thousand are daily active versus seats that got turned on. Muximus: Is the product real enough to care? Penny: The direction is. Salesforce is bundling agents into the place where sales and service already live, at a predictable seat price. That's easier for finance than a surprise token bill. The question is whether Coworker actually finishes work, or mostly drafts things a person still has to check. Muximus: So what does a Salesforce customer do? Penny: Before you buy seats for the whole company, pick one job with a clear finish line. A renewal brief, a case summary, something you already measure. Run it for two weeks with a small group. Count how often a person still had to rewrite the answer, and how often it actually moved the deal or the ticket. Then decide if fifty dollars a seat is cheaper than the time it saves. Muximus: Thanks, Penny. Now, quick hits. Penny: Meta has acquired Conductor, an A I coding startup that runs parallel coding agents. Meta confirmed the deal. Conductor's cloud service shuts down on November twentieth, and cloud data gets deleted after that, so anyone using it in the cloud needs a plan this month. Muximus: OpenAI says it will acquire Pierre Computer Company, the team behind Code Storage and open source tools already used inside Codex for browsing files and reviewing diffs. Pierre's open source projects stay open. The deal still needs regulatory approval. Penny: Google's developer blog argues that price per million tokens is the wrong scoreboard for agents. Measure cost per successful task, and put a cheap classifier in front so easy work doesn't hit an expensive thinking model. Useful framing even if you never touch their sample router. Muximus: And OpenAI launched Images two point six in ChatGPT, Codex and the A P I, with faster generation and animated G I Fs. Handy for marketing teams. Worth a line in your brand rules about what staff are allowed to generate and publish. Muximus: Okay, so that's the news. Today we talked about Uber's tiny A I agent booking numbers, Apple putting daily caps on staff A I spend, and Salesforce pitching Coworker as a seat priced teammate inside the C R M. Penny: It's the gap between the pitch and the meter. Agents that barely book a ride, budgets that suddenly have ceilings, and a vendor asking for fifty dollars a seat before you've proven the job. Muximus: Three different ways to ask for receipts. Penny: Pretty much. AI Squawk is produced and brought to you by Var Ops. If today's stories made you want clearer proof before the next A I purchase, that's the work we do. Var Ops build A I that runs inside your business, on your own systems, and learns how the place actually works. So the work stays in your house, and so does your data. Start at Var Ops dot com. Muximus: If this helps you push back on A I hype with better questions, follow the AI Squawk podcast wherever you're listening, so you don't miss an episode. And for more columns from Nix, built on real data and hard-earned lessons from the trenches, sign up for Var Ops newsletters. Muximus: That's it from us for today. Thanks for spending a few minutes with us. Penny: Enjoy the rest of your Thursday, everyone. We'll see you tomorrow. Muximus: Take care.