Hey, curious people.
Uber’s CFO told investors how many rides AI agents booked in Halloween week. You can count them on one hand and keep your thumb free.
On the menu this week:
🖥️ Uber counted its AI agent bookings. Four rides.
🛠️ The play: analysts followed an agent’s hidden lean 86% of the time. Build in the counterargument.
🔍 Higgsfield: campaign two decides the renewal.
📡 Radar: AI stocks sink the S&P 500 4.4%, Claude gets a Personal Finance beta, Airbnb builds its own agent, and six more.
01 / THE BIG MOVE
🖥️ AI agents booked four Uber rides
So, when are agents supposed to start becoming buying agents? At least for Uber, the answer is not yet.
Uber counted. In Halloween week, its busiest of October, AI agents booked four rides and about 500 Uber Eats orders, per TechCrunch.
The number that matters: 4 rides, out of about 320 million trips and orders that week, per CFO Balaji Krishnamurthy.
UBER’S HALLOWEEN WEEK
AGENT RIDES
4
AGENT EATS ORDERS
~500
ALL TRIPS AND ORDERS
~320M
Uber’s figures from its Q3 call, via TechCrunch.
What the CEOs said:
Uber: “very, very early and very limited traction,” per Dara Khosrowshahi, who still expects agents to add demand. He says a ride takes about 50 seconds to book in the app. Tough bar for a robot.
DoorDash: “maybe several hundred orders in total” from third-party agents, per Tony Xu, via TechCrunch.
Amazon: “still not great,” per Andy Jassy, via Fortune. Then again, Amazon blocked Meta’s Muse from its store.
Why it matters for GTM: the agent that sells is real. I ran six chats with Anthropic’s, and it talked me up to Enterprise.
The agent that buys is still a rounding error. My bet: buyers lean on agents for research long before they let one pay.
Count agent orders in your own logs before anyone asks for budget. Uber’s fit on a sticky note.
02 / PUT IT TO WORK
🛠️ The play: build the counterargument into the workflow
Anthropic handed 50 analysts an agent with a secret lean. A coin flip picked which option it favored (the most honest A/B test I’ve seen all year).
How often analysts went the agent’s way
WHAT YOU’D EXPECT FROM A COIN FLIP
CHOICES THAT FOLLOWED THE AGENT’S LEAN
CREDITED THE AGENT FOR THEIR ARGUMENTS
They sided with the agent almost twice as often as chance. Only 15% gave it credit.
Anthropic, 50 professional analysts using Claude Code, October 2026.
Anthropic’s numbers, and it turned the lean up on purpose. One idea it floats: have the agent make the best case for each option.
Where it fits: nearly every business call can be argued both ways. Which accounts to work, whether to discount, what to test next. A confident agent picks a side, and a smart analyst never notices.
Take the analyst out and the lean becomes the plan. No one’s left in the room to say “wait.”
So the fix lives in the workflow: a second pass that argues the other side before anyone acts. In an agentic pipeline, that’s its own step.
TRY THIS ON ONE TASK
Add a devil’s advocate step to one AI decision.
Pick one recommendation your team gets from AI on a schedule. Lead scoring, discount approvals, this week’s outreach list.
Add a step that sees only the recommendation and argues against it. If it wins too often, the first step was leaning. Fix the prompt. Keep the checker.
03 / GROWTH TEARDOWN
🔍 Higgsfield: the second campaign decides renewal
Higgsfield makes AI video ads. Nvidia and TCV extended its Series B last month, per Bloomberg, and it says it has passed 40 million registered users.
THE SHORT VERSION
Why people buy: you see a finished ad before you write a prompt.
Where growth comes from: pay-as-you-go credits, now nearly 40% of revenue, up from under 10% (my arithmetic on Higgsfield’s figures).
What I’d test: a package that gets a small agency through campaign two without calling support.
Disclosure: I work at Canva, which also makes creative tools. So this was interesting to me.
04 / PIPELINE SIGNALS
📡 On the radar
Wall Street had a rough week. The product teams didn’t notice.
AI stocks led a 4.4% drop in the S&P 500 on Thursday, its worst day since April 2025, on worries about data center moratoriums, per AP.
Anthropic started rolling out Personal Finance in Claude, a US beta that reads up to two years of transactions and, it says, can’t move your money.
Airbnb says its own AI travel agent arrives in the first half of 2027, per TechCrunch.
HubSpot now expects net new ARR to fall in 2026 and took its credit-based go-to-market seat pilot across EMEA, per its CFO.
Meta bought Conductor, which runs several coding agents at once, and Conductor Cloud starts deleting data Nov 20.
OpenAI released Images 2.6 in ChatGPT and Codex on every tier, with saved reference images and GIFs, per TestingCatalog.
ChatGPT’s search-like queries are roughly one-fifth of Google’s, by OpenAI’s own definition, per its Adam Fry.
Anthropic’s amended S-1 shows about $17B in Q3 revenue and about $895B in commitments, per Yahoo Finance.
Anthropic renamed Max 20x to Ultra and dropped the five-hour limit for verified accounts, and OpenAI says ChatGPT and Codex will share one allowance.
OVER TO YOU
💬 Has an AI agent bought anything from your company yet?
Six chats with Anthropic’s agent showed me that selling agents are real today. I’d want to see agent orders in my own logs before I spent a dollar on the buying side. What have you seen?
How was this week’s issue?
Robert
P.S. Know someone trying to put AI to work in go-to-market? Forward this to them. They can subscribe at bizgrowth.ai.