Technology

1. Gulf is still dealmaking
2. Big Tech’s scam problem
3. Bilt CEO on agentic commerce
4. ChatGPT vs Gemini
5. China’s agent hype
6. Anthropic goes to war

What we can learn from the early days of the telephone, and scientists made a functioning digital version of a fly’s brain.

At this point in the AI craze, we’ve moved past the doomsayer proclamation that AI will obliterate all our jobs. We know AI will displace some jobs and create new ones, but the question most CEOs are wrestling with right now is, how long will the lag period — and friction — between these two cycles last? My bet is that it will take longer than we think — let me tell you why.

If past is prelude, a good example of how rapid technological change can upend economies is the invention of the telephone, which, as Semafor’s J.D. Capelouto writes this week, is celebrating the 150th anniversary of Alexander Graham Bell’s first-ever phone call.

No one in my generation remembers the huge labor force of switchboard operators, but in the 1950s there were 340,000 workers physically plugging cords into jacks to connect callers to one another (roughly equivalent to the number of dental assistants today). Those switchboard jobs are a thing of the past now, but they weren’t phased out overnight.

It took at least six decades after the automatic switching system was invented — even though that tech was more than capable of doing the operator’s job — to phase the jobs out.

At the time, phone companies thought making users dial their own phones was too laborious. Switchboard operators (most of whom were women) were viewed as service providers, according to research from the Richmond Fed. The operators got to know many of their customers personally, passed important messages along if the receiver didn’t pick up, and chatted callers up about the weather, sports games, and big news. In other words, it was their human attributes — service, knowledge, and flexibility — that kept them useful.

Fast forward four years into the AI boom, and the human attributes workers bring are more important than ever. Even though Claude can offer recommendations on how to deal with, say, a sensitive client issue, I still seek out my (human) boss’ advice.

Eventually, telecom operators succumbed to automation, largely due to financial pressure and software development, but it took a while. And if history continues to repeat itself, today’s working-age employees will mostly hang on to their jobs. Their kids, however, will be a different story.

War doesn’t stop deals in the Gulf

The biggest dealmakers in the AI-hungry Gulf aren’t slowing down because of the war in Iran — and a number of them are announcing deals in a deliberate bid to show strength. Abu Dhabi’s Judan, the new $237 billion financial services holding company helmed by Deputy Ruler Sheikh Tahnoon bin Zayed, is buying a majority stake in Alpha Wave Global, a Miami-based alternative asset manager and investor in Anthropic, OpenAI, and SpaceX.

Sheikh Tahnoon’s sprawling conglomerate, International Holding Co., created Judan last month, consolidating its banking, insurance, investment, and financial technology interests under one firm.

Last week, Qatar Investment Authority bought into US clean energy supplier AES, and Aluminium Bahrain, controlled by the kingdom’s sovereign fund, agreed to acquire the EU’s biggest aluminum smelter.

Right now, the appetite to show that life carries on as usual by continuing to do outbound deals is high, Semafor’s Matthew Martin wrote. But it will be more difficult to press the case that it’s business as usual for attracting money into the region.

Big Tech faces big scam problem

Tech companies announced a slew of new security and scam-prevention tools this week, in their ongoing efforts to clean up the AI slop — and security risks — their technology helped create.YouTube expanded its deepfake detection technology to politicians and journalists, while Meta is now warning Facebook users about suspicious friend requests and when potential scammers appear to try linking a user’s WhatsApp account to their device.

Meta’s tools are “part of a multilayer defense,” and are not sufficient on their own, Meta’s David Agranovich, director of global threat disruption, told reporters. That’s because scammers are increasingly operating across multiple platforms and using AI to make their accounts more realistic. The nature of cybersecurity means tech companies are largely responding to scams, rather than getting ahead of them. And as AI helps bad actors with ever more sophisticated tactics, tech companies will continue to struggle to put the cat back in the bag.

— Rachyl Jones

Bilt’s bet on agentic commerce

Building up. Bilt Rewards, which gained popularity for letting users pay their rent with a credit card and earn rewards for it, is now betting it can become a membership club that gamifies spending across local stores and restaurants. The company recently launched a “Neighborhood Concierge” chatbot that helps manage adulting tasks like getting a leak fixed, paying monthly bills, or making restaurant reservations.

Right now, it’s too hard to get AI to replace all of that. So the company is stepping in as a sort of middleman.

“The best AI in the market today gives you a lot of information, but it’s not hooked into the ecosystem around it yet,” CEO Ankur Jain said on the latest episode of Semafor’s Compound Interest show. “A lot of these businesses would not let a third-party AI come in and just mess with their back end.”

The kind of integration Bilt is aiming for, which relies on partnerships with local businesses, is the picture of an agentic world.

AI as a gatekeeper between brands and buyers

Companies are now writing ad copy for bots, not people, and those bots are telling consumers what products to buy. Google’s AI Overviews and OpenAI’s ChatGPT are starting to portray the same brands differently: Google is 44% more likely to skew negative about a given brand than ChatGPT, serving up criticism around lawsuits, product recalls, and news, according to a report by SEO platform BrightEdge. When ChatGPT goes negative, it does so around product reviews and costs.

With fewer humans in the process and AI that can sum up an entire brand’s history into one recommendation, some companies have taken to generating a slew of positive content in an attempt to rank better in AI results. That may help advertisers, but it floods the internet with a kind of product slop that could also dissuade consumers. The rules of advertising are changing, with SEO becoming a thing of the past. It’s likely the new standards will remain in flux as AI gets better and shoppers change their habits.

— Rachyl Jones

AI agent hype booms in China

AI agents are the hottest new thing in China’s tech scene. OpenClaw, which can execute tasks autonomously on a user’s computer using AI models, has taken off in the AI-optimistic country: China’s tech giants have launched their own OpenClaw-like AI agent products, and around 1,000 people — ranging from primary school students to retirees — recently lined up outside Tencent’s headquarters for a free installation event. Local governments in Wuxi and Shenzhen have rolled out policy measures to support OpenClaw development, even as state media issued a security warning over the agents. “This is how China scales technology,” a tech analyst wrote. “Government sets the table with money and policy. Companies bring the products.”

Anthropic sued the Pentagon: now what?

Anthropic’s Monday lawsuit against the Pentagon, which designated it as a supply chain risk for the US, provides an illustrative look into the future of AI and war. Anthropic argues that the government applies the statute incorrectly, overextends the executive branch’s authorities, violates due process, and infringes on its free speech protections.

Legal experts say Anthropic has a strong case because the label — historically wielded against Chinese and Russian companies — is meant to prevent outside sabotage and spying, which Anthropic isn’t accused of.

In a filing in support of Anthropic, Microsoft also fired off a warning to the US government not to mess with AI companies. Urging the court to issue a temporary restraining order that would block the “supply chain risk” designation for Anthropic’s existing contracts, it warned “this could potentially hamper US warfighters at a critical point in time.”

This April, Anne Wojcicki, Founder and CEO of 23andMe Research Institute, will join global leaders at Semafor World Economy — the premier convening for the world’s top executives — to sit down with Semafor editors for conversations on the forces shaping global markets, emerging technologies, and geopolitics. See the first lineup of speakers here.

Scientists used a digital model of a fly’s brain to control a digital fly. In 2024, an international project mapped a fruit fly’s brain; now, a startup has used that map in a neuron-for-neuron digital model, and installed it in a simulated body in a simulated world. The digital fly demonstrated fly-like actions. The simulation is incomplete — the researchers treated all neurons as basic and identical, when they are complex and varied. The wider nervous system was not scanned, so researchers had to guess