ChatGPT Can Now Shop, Research, and Work For You (While You Sleep)
4IR - Daily AI News
Welcome back to 4IR. Here's today's lineup:
OpenAI transforms ChatGPT into AI agent that handles complex tasks using its own virtual computer
Google DeepMind's Gemini wins gold at Mathematical Olympiad, solving 5 out of 6 problems within time limit
Trump's "anti-woke AI" executive order mandates federal AI procurement compliance
Tools & Trending
🔥 TOP STORY: ChatGPT Agent launches as OpenAI's boldest AI assistant yet
The story: OpenAI unveiled ChatGPT Agent on July 17, transforming its chatbot from conversation partner to autonomous task executor. Unlike previous AI assistants that only respond to prompts, Agent Mode gives ChatGPT its own virtual computer to navigate websites, analyze data, and complete multi-step workflows from start to finish. The system can handle requests like "look at my calendar and brief me on upcoming client meetings based on recent news" or "analyze three competitors and create a slide deck" — all while keeping users in control with permission prompts for important actions.
What we know:
Agent Mode unifies three breakthrough capabilities: Operator's web navigation, Deep Research's synthesis skills, and ChatGPT's conversational intelligence
It runs on a virtual machine with visual browser, text browser, terminal access, and API connections to Gmail and GitHub
Available to Pro ($200/month), Plus ($20/month), and Team subscribers through the tools dropdown, with Enterprise coming in weeks
Why it matters: This isn't just another ChatGPT feature — it's OpenAI's attempt to fulfill the decade-old promise of AI agents that actually work. Internal testing shows Agent significantly outperformed both Deep Research and o3 models at complex tasks. CEO Sam Altman positioned it as "cutting edge and experimental" but warned against high-stakes use until the system proves itself in the wild.
The timing is strategic. As Google races to deploy Gemini agents and Microsoft pushes Copilot integration, OpenAI is betting that autonomous task execution — not just better chat — will define AI's next phase.
🏛️ POLICY SHIFT: Trump's "anti-woke AI" executive order reshapes federal procurement
The story: On July 23, President Trump signed an executive order mandating that all federal AI contractors certify their systems are "free of ideological bias" to remain eligible for government contracts. The "Preventing Woke AI in the Federal Government" order specifically targets AI systems that incorporate "critical race theory, transgenderism, unconscious bias, intersectionality, and systemic racism" — marking a dramatic reversal from Biden-era policies focused on preventing AI bias against marginalized groups.
What we know:
Federal contractors must now prove their AI systems don't "suppress or distort factual information about race or sex" or "manipulate racial/sexual representation in outputs"
The General Services Administration will create new procurement language requiring AI models to meet "standards of truthfulness" as a contract condition
Definitions remain deliberately vague — White House officials couldn't specify how "ideological bias" would be measured, leaving contractors to interpret compliance
Why it matters: This isn't just federal policy — it's a cultural battleground with billion-dollar stakes. Major AI companies from Microsoft to Google now face a choice between maintaining current safety guardrails or accessing lucrative government contracts. The order comes as tech leaders like Elon Musk's Grok faced criticism for extremist outputs, yet still secured defense contracts. The timing is strategic: China simultaneously launched its Global AI Action Plan, positioning itself as the inclusive alternative to America's ideological AI restrictions.
The broader question: Can we align AI with any human values when the definition of "truth" itself becomes political?
🎯 REALITY CHECK: Mathematical AI breakthrough masks growing safety concerns
What happened: Both Google DeepMind and OpenAI achieved gold-medal scores at the 2025 International Mathematical Olympiad, each solving 5 out of 6 problems and scoring 35 out of 42 points. DeepMind's Gemini Deep Think operated entirely in natural language within the 4.5-hour time limit, while OpenAI's model matched the performance in parallel testing. IMO President Gregor Dolinar called the solutions "astonishing in many respects" and praised their clarity and precision.
The concerning reality: OpenAI simultaneously warned that ChatGPT Agent poses "high" bioweapon risks, marking the first time the company has classified a consumer product with this danger level. Safety researchers found the model could provide "meaningful assistance to novice actors" in creating biological threats. The same mathematical reasoning that earned gold medals could guide someone through complex chemical synthesis or biological engineering.
Why this matters: We're witnessing AI systems achieve human-level performance in abstract reasoning while simultaneously becoming more dangerous. The models that solve elite mathematics problems are now powerful enough to potentially aid bioweapon development. As one safety researcher noted: "Unlike nuclear threats, obtaining materials is less of a barrier for bio threats — security depends on scarcity of knowledge, which AI is rapidly eroding."
🛠️ HOW-TO: Master ChatGPT's new Agent Mode for complex workflows
What you'll build: An AI assistant that handles end-to-end business research and presentation creation.
Steps:
Access Agent Mode in ChatGPT Pro, Plus, or Team by clicking the tools dropdown and selecting "agent mode"
Start with a complex, multi-step prompt: "Research the top 3 competitors in the electric vehicle charging market, analyze their pricing models, create a comparison spreadsheet, and generate a 10-slide presentation summarizing key findings"
Let Agent work autonomously — it will use its virtual browser to research websites, prompt you to log into data sources when needed, and switch between analysis and action
Monitor the process through the visual interface showing what Agent is doing on its virtual computer — you can interrupt, redirect, or approve actions at any time
Pro tip: Agent excels at tasks requiring multiple tools — combine calendar analysis with web research, or data gathering with document creation. Start with lower-stakes tasks to understand its capabilities before handling sensitive information.
⚡ QUICK HITS
New Tools
ChatGPT Agent — OpenAI's unified task automation system for Pro/Plus/Team users
Gemini Deep Think — Google's mathematical reasoning mode achieving IMO gold standard
Google AI Ultra — New subscription tier at $249.99/month with highest usage limits
Veo 3 video generation — Now available in Gemini app for AI Ultra subscribers
Other News
Trump administration launches AI Action Plan with 90+ federal policy actions
Microsoft mandates AI tool usage for all employees, making GitHub Copilot adoption required
5 human competitors achieved perfect IMO scores vs. AI's 35/42 gold-medal performance
Ford CEO warns AI will replace "half of all white-collar workers" as industry leaders acknowledge job displacement