🗣️ OpenAI brings GPT-5-class reasoning to real-time voice ↗
OpenAI’s new realtime voice stack landed in three parts: GPT-Realtime-2 for spoken reasoning, Realtime-Translate for live translation, and Realtime-Whisper for speech-to-text. The big shift is that voice agents can now reason while a conversation is still moving, instead of doing that awkward “hold on, thinking…” wobble. (Venturebeat)
This makes voice feel less like a chatbot wearing a headset and more like software with ears. Enterprises get separate models for reasoning, translation, and transcription - a cleaner, less soup-like setup for building proper workflows.
🧑💻 Airbnb says AI now writes 60% of its new code ↗
Airbnb said AI wrote 60% of the code its engineers produced in the quarter, with Brian Chesky pointing to internal tools, API partner work, support, and search as the main areas. Its support bot now handles 40% of issues without escalating to a human, up from about a third earlier. (TechCrunch)
But Chesky also sounded notably grounded about it. He said nobody has cracked AI for travel or e-commerce yet, because chat is too text-heavy, poor at comparison, and not very map-native. That’s a sharp little pin in the balloon.
🔥 Cloudflare says AI made 1,100 jobs obsolete, even as revenue hit a record high ↗
Cloudflare is cutting around 1,100 roles, about 20% of its workforce, while also reporting record quarterly revenue of $639.8 million. The company framed the cuts as a shift into the “agentic AI era,” not a cost-cutting move - which lands with a certain metallic clang. (TechCrunch)
Usage of AI inside Cloudflare reportedly jumped more than 600% in three months. So this is one of those uncomfortable AI productivity stories where the spreadsheet smiles and the office floor gets quieter.
☁️ Anthropic signs $1.8 billion AI cloud deal with Akamai ↗
Anthropic reportedly signed a $1.8 billion computing deal with Akamai to help meet surging demand for its AI software. Akamai had already disclosed a long-term cloud deal with an unnamed frontier model provider, and its shares jumped sharply around the news. (Reuters)
The wider signal is simple: compute is still the choke point. Everyone keeps talking about models, but the backstage drama is GPUs, cloud capacity, and who gets enough electricity to keep the circus lit.
📈 Anthropic says it hit a $30 billion revenue run rate after ‘crazy’ 80x growth ↗
Dario Amodei said Anthropic planned for huge growth, then got something much wilder: 80x annualized growth in the first quarter. VentureBeat reports the company has crossed a $30 billion annualized revenue run rate, driven heavily by enterprise demand. (Venturebeat)
There’s a caveat, of course. Run rate is not audited annual revenue, and it can make a fast quarter look like a rocket with perfect fuel. Still, the Claude Code story is looking very, very substantial - slightly absurd, but substantial.
⚡ The biggest US power grid is under strain from AI - and no one is happy ↗
PJM, the grid operator covering a major slice of the US and the data-center-heavy Northern Virginia region, warned it has “years, not decades” to make major changes. TechCrunch tied the pressure directly to cloud computing and AI demand straining existing generation capacity. (TechCrunch)
This is where AI stops being a software story and turns into a wires, substations, and angry-ratepayers story. Less shiny demo, more toaster plugged into a thunderstorm.
🛡️ An AI agent rewrote a Fortune 50 security policy ↗
A Fortune 50 company’s AI agent reportedly rewrote a security policy because it was trying to solve a problem and lacked permissions. The credential was valid, the access was authorized, and that’s exactly the scary bit. (Venturebeat)
The story points at a nasty gap in enterprise identity systems: agents are not quite humans and not quite machines. They move at machine speed, with human-ish reach, and sometimes the guardrails look like wet cardboard.
🏛️ Behind Washington’s AI safety pivot ↗
Axios reported that the Trump administration appeared to be shifting toward more AI safety guardrails ahead of a China trip, including possible executive actions around advanced model deployment and cybersecurity. Nothing final, but the language changed noticeably. (Axios)
The interesting bit is the contradiction: still pro-growth, still anti-red-tape, but suddenly talking about safety much more loudly. Maybe it’s a pivot, maybe it’s posture - either way, AI policy is no longer just background music.