Sunday, August 3, 2025

AI InStone Newsletter

This month’s dive into AI reflects the chaotic frontier-like AI marketplace changes that are happening if not weekly, then daily – mid-summer heat is tracking AI development. Across Silicon Valley and beyond, the world's leading AI companies have unveiled transformative technologies that signal we are entering what can only be described as the "agentic era" of AI—where artificial intelligence systems don't just respond to queries but actively reason, plan, and execute complex tasks autonomously.

From the Major Players

OpenAI's Bold Leap into Autonomous Action

OpenAI has made perhaps the most significant announcement with the introduction of ChatGPT Agent on July 17, 2025, described as "bridging research and action." OpenAIOpenAI This represents a fundamental shift from conversational AI to autonomous action. ChatGPT can now "do work for you using its own computer, handling complex tasks from start to finish," with capabilities spanning from analyzing competitors and creating slide decks to planning meals and making purchases. Introducing ChatGPT agent: bridging research and action | OpenAI

The technical achievement here cannot be overstated. At the core is "a unified agentic system" that brings together three major breakthroughs: Operator's ability to interact with websites, deep research's skill in synthesizing information, and ChatGPT's intelligence and conversational fluency. Introducing ChatGPT agent: bridging research and action | OpenAI Users can now request complex multi-step tasks like "look at my calendar and brief me on upcoming client meetings based on recent news" and watch as the AI autonomously navigates the web, analyzes information, and delivers comprehensive results.

Meanwhile, the highly anticipated GPT-5 release continues to build momentum. Sources close to OpenAI indicate GPT-5 is expected to launch in August 2025, with CEO Sam Altman recently posting that it would be released "soon." OpenAI prepares GPT‑5 launch for August 2025, sources say GPT-5 is designed to "unify several capabilities under one system" and is expected to combine traditional language model attributes with advanced reasoning capabilities like those seen in the o3 series. ChatGPT-5 launch looks imminent — here’s everything we know so far | Tom's Guide

Google's Multi-Agent Revolution with Gemini Deep Think

Google has made its own dramatic entrance into advanced AI reasoning with the launch of Gemini 2.5 Deep Think on August 1, 2025. This represents "Google's first publicly available multi-agent model" where "multiple AI agents tackle a question in parallel." Google rolls out Gemini Deep Think AI, a reasoning model that tests multiple ideas in parallel | TechCrunch Unlike traditional sequential processing, this system spawns multiple AI agents simultaneously to explore different approaches to complex problems.

The technical breakthrough is significant: "Google used a variation of Gemini 2.5 Deep Think to score a gold medal at this year's International Math Olympiad (IMO)." Google rolls out Gemini Deep Think AI, a reasoning model that tests multiple ideas in parallel | TechCrunch While that specialized model "takes hours to reason," the consumer version is faster and more practical for daily use. Deep Think achieves "state-of-the-art performance across LiveCodeBench V6, which measures competitive code performance, and Humanity's Last Exam, a challenging benchmark that measures expertise in different domains." Gemini 2.5: Deep Think is now rolling out

The implications extend far beyond academic benchmarks. Google claims the model could "aid researchers and potentially accelerate the path to discovery" through its enhanced reasoning capabilities. Google rolls out Gemini Deep Think AI, a reasoning model that tests multiple ideas in parallel | TechCrunch This positions Google as a serious competitor in the race toward artificial general intelligence, with multi-agent collaboration as a key differentiator.

A quick note of Google’s “other” projects – AlphaEarth Foundations and Google Earth Engine uniting petabytes of observational data with new AI models and influencing partners in their own initiatives for building real-world insights and solutions.

Perplexity's Browser Revolution and Market Expansion

Perplexity has continued its aggressive expansion with significant developments across multiple fronts. In July 2025, Perplexity launched Comet, an AI browser based on Chromium, initially available to users subscribed to the highest tier, with broader availability expected over time. Perplexity AI - Wikipedia This represents a direct challenge to traditional web browsing by integrating AI-powered search and analysis directly into the browser experience.

The company's growth metrics are staggering. Perplexity AI has reached approximately 2 million daily visitors worldwide, and in India alone, users increased by 640% year-over-year in Q2 2025. Perplexity AI Statistics 2025 – MAU & Revenue (Users Data) The company recently secured a $100 million funding round that boosted its valuation to $18 billion, reflecting growing investor confidence in AI-powered search alternatives. Perplexity AI Statistics 2025 – MAU & Revenue (Users Data)

Perplexity has also launched Perplexity Max, its most advanced subscription tier, providing "unlimited Labs usage per month" and priority access to new features. Introducing Perplexity Max This premium offering signals the company's confidence in monetizing advanced AI capabilities as user demand grows.

Apple's Measured AI Integration

Apple has taken a characteristically different approach, focusing on privacy-first AI integration across its ecosystem. At WWDC 2025 in June, Apple announced new Apple Intelligence features coming to iPhone, iPad, Mac, Apple Watch, and Apple Vision Pro, with developers now able to access the on-device foundation model to power private, intelligent experiences within their apps. Apple Intelligence gets even more powerful with new capabilities across Apple devices - Apple

Apple Intelligence features are expanding to eight more languages by the end of the year: Danish, Dutch, Norwegian, Portuguese (Portugal), Swedish, Turkish, Chinese (traditional), and Vietnamese. Apple Intelligence gets even more powerful with new capabilities across Apple devices - Apple The company's strategy emphasizes on-device processing and privacy, distinguishing it from cloud-dependent competitors.

However, some anticipated features remain delayed. The more personalized version of Siri that was expected to understand "personal context" has been postponed, reportedly due to the system being "too error-ridden to ship." Apple Intelligence: Everything you need to know about Apple's AI model and services | TechCrunch This suggests Apple is maintaining its quality-first approach even as competition intensifies.

Anthropic's Focus on Enterprise and Safety

Anthropic has maintained its position as the safety-focused AI leader while making strategic moves in enterprise markets. Based on the provided document, the company has introduced new weekly rate limits for Claude Code subscribers and launched a unified interface specifically for analyzing financial market data in collaboration with major banks. This represents a significant push into specialized enterprise applications where AI safety and reliability are paramount.

Anthropic continues to prioritize interpretable and steerable AI while launching initiatives for labor market impact research. This approach differentiates them in a market increasingly concerned about AI's societal implications.

Microsoft's Agentic Integration Across the Platform

Microsoft has been systematically integrating agentic AI capabilities across its entire ecosystem. On July 28, Microsoft launched "Copilot Mode" in Edge browser, allowing users to browse the web while being assisted by AI that can understand what they're researching, predict what they want to do, and take action on their behalf. Microsoft Edge is now an AI browser with launch of 'Copilot Mode' | TechCrunch

Recent updates to Microsoft 365 Copilot in July 2025 include voice interaction with Copilot Chat on mobile devices, unified conversation history across platforms, and real-time translation capabilities in Teams meetings. Microsoft 365 Copilot August 2025 Updates: Enhanced AI Tools for Admins and Users | Windows Forum The integration extends beyond individual applications to system-wide intelligence.

At Microsoft Build 2025, the company introduced Windows AI Foundry, offering "a unified and reliable platform supporting the AI developer lifecycle across training and inference" and announced Microsoft 365 Copilot Tuning, allowing customers to "use their own company data, workflows and processes to train models and create agents." Microsoft Build 2025: The age of AI agents and building the open agentic web - The Official Microsoft Blog

Meta's Superintelligence Ambitions

Meta has perhaps made the most audacious moves, with CEO Mark Zuckerberg announcing an all-out push toward artificial superintelligence. In June 2025, Zuckerberg created Meta Superintelligence Labs, led by former Scale AI CEO Alexandr Wang as chief AI officer, following a shocking $14.3 billion investment in Scale AI. CNBCCNBC

On July 25, Zuckerberg announced that Shengjia Zhao, co-creator of OpenAI's ChatGPT, will serve as chief scientist of Meta Superintelligence Labs. Meta names OpenAI's Shengjia Zhao as chief scientist of AI Superintelligence Lab The company's recruiting strategy has been aggressive, with reports of "pay packages worth hundreds of millions of dollars to new AI hires" and OpenAI CEO Sam Altman claiming Meta is offering his employees "$100 million signing bonuses." Meta is shelling out big bucks to get ahead in AI. Here’s who it’s hiring | CNN Business

Zuckerberg announced plans to invest "hundreds of billions of dollars" into AI compute infrastructure, with the first supercluster, Prometheus, coming online in 2026. Meta CEO Zuckerberg says first AI data supercluster will come online in 2026 This massive investment underscores Meta's commitment to competing at the highest levels of AI development.

Policy, Tools, and Technology

AI Action Plan (U.S.)

  • White House Policy Launch: The Trump administration released its long-awaited AI Action Plan on July 23, centering on accelerating AI innovation, expanding U.S. infrastructure, and asserting global technology leadership.
  • Executive Orders: Accompanying orders targeted AI data center buildout, secure AI exports, and requirements for “unbiased” AI in government use. The plan aims to streamline regulations, spur open-source and open-weight AI development, and embed AI benefits in the workforce1415.
  • Bipartisan Task Force: A new national AI task force will align policy across sectors like education, defense, and workforce to support the plan’s recommendations16.
China’s AI Plan
  • Global AI Governance Proposal: China rapidly answered the U.S. plan at the World AI Conference (WAIC), pitching a global AI governance framework to counter U.S. “go-it-alone” strategy. The Chinese plan focuses on international coordination, safety standards, and counteracting technological monopolies1718.
  • Strategic Focus: China underscored its intent to lead on ethics and cross-border AI regulation, stressing the need for AI safety and government scrutiny of commercial systems18.

Stanford's Scientific AI Innovation

Stanford researchers have been pushing the boundaries of AI applications in scientific research. Stanford Medicine researchers created a team of "virtual scientists" backed by artificial intelligence to help solve problems in their real-world lab, complete with an AI principal investigator and seasoned scientists. Stanford MedicineStanford University

The virtual lab has already demonstrated its potential by tasking the "team" to devise a better way to create a vaccine for SARS-CoV-2, accomplishing this in just a few days. Researchers create ‘virtual scientists’ to solve complex biological problems | Stanford Report This represents a fascinating application of multi-agent AI systems to accelerate scientific discovery and research processes.

Understanding What it Takes to be an AI PC

PC Magazine does a good job of exploring these waters and the hardware nuances needed to run AI locally on a PC in their article entitled, “What Is an AI PC? How AI Will Reshape Your Next Computer,” by Brian Westover. This article is already “dated” from the first half of 2025, but describes a PC configuration and its respective hardware, software and bios changes needed to run dedicated AI agents.

Technology Trends – Generally Speaking

McKinsey’s Technology Trends outlook for 2025 is a fascinating dive into the global technology landscape – this issue focusing on “agentic AI,” but looking a wide range of so-called “frontier technologies.”

Editorial: The Dawn of the Agentic Era

These developments collectively represent more than incremental improvements—they signal a fundamental transformation in how we interact with artificial intelligence. We are witnessing the emergence of what can only be called the "agentic era," where AI systems transition from reactive tools to proactive partners capable of autonomous reasoning, planning, and execution.

The convergence is striking: OpenAI's ChatGPT Agent can autonomously navigate websites and complete complex tasks; Google's Deep Think employs multiple AI agents working in parallel; Meta is betting hundreds of billions on superintelligence; and Microsoft is embedding agentic capabilities throughout its platform. Even Apple, with its measured approach, is moving toward more contextual, action-oriented AI.

This shift raises profound questions about the future of work, creativity, and human agency. When AI systems can autonomously research, analyze, plan, and execute complex tasks, what remains uniquely human? The answer, increasingly, seems to be setting objectives, providing context, and making value judgments about outcomes.

The competitive dynamics are also fascinating. While OpenAI and Google race toward more powerful reasoning models, Meta is pursuing superintelligence through massive compute and talent acquisition. Apple focuses on privacy-first integration, and Microsoft leverages its platform dominance to make AI ubiquitous. Each approach reflects different philosophical assumptions about AI's role in society.

Perhaps most significantly, we're seeing the emergence of AI systems that can collaborate—not just with humans, but with each other. Google's multi-agent Deep Think and Stanford's virtual lab scientists suggest a future where complex problems are solved by teams of specialized AI agents working in concert.

The next 12-18 months will likely determine whether these agentic AI systems fulfill their promise of augmenting human capability or whether they introduce new risks and challenges we haven't yet fully understood. What's certain is that the AI landscape of late 2025 would be barely recognizable to observers from just two years ago. We are living through a transformation as significant as the invention of the internet itself, compressed into a timeframe that would have seemed impossible just a decade ago.

The age of agentic AI has begun, and its implications will ripple through every sector of human activity in the months and years to come.

Sources (GenAI generated, mostly)

No comments:

Post a Comment