☕️ OpenAI's General-Purpose "ChatGPT Agent" Enters The Fray

With general-purpose AI agents driving the latest automation trend in the industry, OpenAI has joined the race at the frontier with its own.

Here are our main key takeaways:

Join us at AI Tangle as we untangle this week's happenings in AI!

Scale your AI capabilities with vetted engineers, product managers, and builders - delivered with enterprise rigor. Athyna offers:

OpenAI launches its latest general-purpose agent - here's everything you need to know

Following the usual OpenAI X/Twitter tease, the company debuted the rather innocuously named "ChatGPT agent" during a Thursday livestream featuring CEO Sam Altman. Combining the best of its other tools, specific capabilities from Operator and Deep Research, OpenAI's agent can take the reins of your computer and perform tasks ranging anywhere from creating presentations at work to planning parts of your personal life.

​

How does it work, and is it any good?

The ChatGPT agent leverages Operator's action-taking remote browser and Deep Research's ability to synthesize information from the web, and is powered by a specialized, unnamed model. It can also use ChatGPT connectors to access apps like Gmail and GitHub to gather extra information for your prompts. This includes more robust new safety guardrails that monitor the agent in real-time, on top of disabling the memory feature to "avoid misuse."

Though it may be early to tell, early test runs, reviews, and reception from the general userbase seem mixed - some underline the demo as "underwhelming" and the agent taking far too long in practice, while others say it fits power users like a glove. In terms of benchmarks, the agent scores 41.6% on Humanity's Last Exam, roughly twice as high as o3 and o4-mini and just shy of Grok 4's 44.4%. On FrontierMath, one of the hardest math benchmarks, the agent scores 27.4%, 4x as much as OpenAI's previous SOTA score of 6.3% by o4-mini.

Starting July 17, OpenAI's agent will be available to Pro, Plus, and Teams subscribers (excluding EU users) via "agent mode" in ChatGPT's dropdown menu.

📊 Demand Isn't Slowing Down - TSMC's Q2 profits are in, and the company's record-high profit surge of 61% retains the massive demand for AI chips.

🧠 I Think $12 BIllion - Ex-OpenAI CTO Mira Murati has officially secured a $2B seed round for Thinking Machines Labs, a secretive AI startup now worth $12B.

⚡ A Power-Up For AI Mode - Google's latest update lets Gemini make phone calls for you and makes Gemini 2.5 Pro the default for AI Mode.

🔎 Even Better Vision- Despite controversy, Microsoft doubles down in an update to Copilot Vision that now lets the AI see your entire desktop screen.

🔭 Four More Join Meta - Another set of new hires join Zuckerberg's newly-formed AI lab as Meta poaches two AI researchers from OpenAI, two from Apple.

Donald Trump unveils a $70 billion AI and energy plan for Pennsylvania during a summit in Pittsburgh

US President Donald Trump joined oil and tech executives at Carnegie Mellon University to announce $70 billion in AI and energy investments for Pennsylvania, with an additional $92 billion pledged by 20 other participating firms. Hosted by Senator Dave McCormick, the summit touted fossil fuels as the power source for AI, despite many tech leaders stating that the future is renewable. Environmental groups denounced the move en masse, warning of higher bills and pollution, while critics pointed out that AI data centers already consume electricity in quantities higher than most heavy industries and urged a shift to cleaner energy.

​

Netflix' "The Eternauts" marks the media streaming platform's first movie to use generative AI

Netflix has, for the first time, integrated generative AI into one of its productions namely Argentina's sci‑fi drama "El Eternauta" (or "The Eternauts"). Ted Sarandos, co-chief of Netflix, said that the company and its VFX artist partners used AI-powered tools to create a sequence of a building collapse 10 times faster than using traditional tooling. Sarandos pointed out that AI allowed Netflix and its partners to achieve high‑quality effects at a fraction of the usual cost and time, making traditionally big‑budget spectacles much more affordable for smaller studios.

​

Scale AI announces a 14% cut to its workforce shortly after Meta's multibillion-dollar investment

Just weeks following Meta's $14.3 billion investment in the startup and the hiring of CEO Alexandr Wang, Scale AI has confirmed a 14% cut to its 1,400-strong workforce and is ending work with another 500 contractors. In an internal memo released on Wednesday, interim CEO Jason Droege stated that Scale AI had ramped up its generative AI capacity "too quickly" and created "excessive bureaucracy" as a result - the layoffs intend to make the company "nimble" again. Despite this, Droege says the company will reverse course and instead increase its headcount in the second year across its business units.

​

Mistral debuts its inaugural family of open source AI audio models for business

France's AI unicorn Mistral has unveiled Voxtral, its first-ever family of open source AI audio models aimed at enterprise use. Mistral claims that Voxtral delivers production‑grade speech intelligence at "less than half the price" of closed systems, offering transcription, summarization, and real‑time action capabilities powered by Mistral Small 3.1 as its backbone. The lineup includes Voxtral Small for large‑scale deployments, Voxtral Mini for on‑device usage, and a budget‑focused Mini Transcribe API optimized for transcription tasks. Developers can try their hand at integrating Voxtral via Hugging Face or Mistral's Le Chat.

​

Adobe Firefly now allows you to generate realistic audio from sound effect cues

​A recent update to Adobe's Firefly app now offers a beta "Generate Sound Effects" feature that turns simple sound effect voice recordings like "clip clop" into realistic audio synced to video timelines. Users can record onomatopoeic cues alongside footage, such as hoofbeats on pavement, and Firefly will produce multiple effect options to choose from. Meanwhile, Firefly's text‑to‑video gains Composition Reference for matching uploaded video framing, Keyframe Cropping to set start and end frames, and new visual style presets. Adobe stated it plans to extend similar controls to even third‑party AI models in future updates.

​Preplink - Prepare better for your next interview with Preplink, a full-stack job prep platform offering a real-time AI interview copilot, smart mock interviews, and resume optimization.

​Fotor - Fotor is an AI-enhanced online photo editor with features to enhance and edit photos effortlessly - experience simple online photo editing at your fingertips.

​Parafact - Factcheck any human or AI-written text in real-time with Parafact, an AI-powered tool that tracks down reliable sources and citations - all in one click.

What is it Like to Work at OpenAI? (12-min read)

Former OpenAI engineer Calvin French-Owen took a moment to reflect on his stay at OpenAI in a recent blog post. In it, he explores how OpenAI still functions like a fledgling startup, how Codex came together in a seven-week crunch, and talks about one of the biggest misconceptions about the company.

📻 AI Confidential Podcast - Confidential Computing Summit 2025: Day 1 Recap & Interviews​

🔮 The Artificially Intelligent Enterprise - Five Prompting Methods Every AI User Should Know​

💡 AI CIO - There Is No Moat​

🎯 The AI Marketing Advantage - OpenAI’s Next Big Model Is On Hold​​​​​​

Has AI Tangle helped you think more clearly about AI in your business? If so, we'd appreciate a quick testimonial! It helps others understand the value - and helps us improve what we publish each week with every edition.

👉 Leave a short review here!​​​