Another week, another demo of just how deeply AI is weaving itself into our work lives. This time, the numbers and product launches don’t lie—AI isn’t just automating repetitive tasks, it’s surfacing everywhere from legal teams to online shopping, and scaling behind the scenes at a mind-boggling pace. When a billion people use something every week, the ground is shifting.
TOP STORIES
We Now Know How People Really Use ChatGPT for Work—Across 100 Countries
- OpenAI released “Signals” data showing how ChatGPT is actually being used for work in 165 different job activities, across nearly 100 countries.
- The dataset is meant for researchers to dig deeper into AI’s impact on labor, with fresh, real-world evidence.
- Although the data is for research, it hints at just how broad AI’s economic reach is becoming.
- Source
“Shop the Look” Gets Smarter: OpenAI and Fanatics Turbocharge Sports Merch Shopping
- Fanatics rolled out a new AI-powered “Shop the Look” feature for fans to discover gear online, thanks to a collaboration with OpenAI.
- Early tests show a 10% jump in engagement compared to the old version, with around 30,000 MLB products included at launch.
- The new feature lets users get faster, more relevant recommendations as fall sports season ramps up.
- Source
ChatGPT’s Back-End Now Serves Over a Billion Users in Milliseconds
- OpenAI has detailed how it rebuilt its storage platform to handle more than 1 billion ChatGPT users, spread across 40 regions.
- By optimizing on Microsoft’s Azure Cosmos DB, the new Habitat platform keeps responses fast and reliable—even as usage explodes.
- It’s a rare glimpse into the hidden engineering that makes massive AI products actually work for everyone.
- Source
OpenAI Boosts Internal Security for Training Its Most Powerful AI Models
- The company says it has “strengthened containment” during training and evaluation of new, frontier-level models.
- That includes using their own strongest AIs to “red-team” (stress test) security boundaries.
- Most of this is behind the curtain—but how OpenAI secures its future models affects all users in the long run.
- Source
OpenAI Partners With Polish Government Institute for AI Safety Research
- OpenAI signed an agreement with Poland’s NASK National Research Institute to work together on AI safety, security, and technical questions.
- The partnership aims to combine international expertise and better understand emerging risks as AI systems scale.
- Don’t expect immediate product changes, but this could shape how global AI safety gets handled.
- Source
Inside OpenAI’s “Forward Deployed Engineering”—Enterprise AI, Explained
- OpenAI explains what its “Forward Deployed Engineering” (FDE) team actually does: hands-on work with enterprises to solve hard, real-world problems.
- The article gives a rare peek into how teams adapt as AI models themselves get better at “one-shotting” use cases.
- For big companies—or skeptics—this is a signal that AI deployment is often about creative problem-solving, not just plugging in a model.
- Source
THIS WEEK IN AI
Look closer at this week’s flood of releases, partnerships, and technical deep-dives, and a pattern emerges: AI is moving out of R&D labs and into the job descriptions—and workflows—of everyday workers. Forget the old hype cycle. Now, we’re finally seeing companies publicize real usage numbers (over a billion weekly users for ChatGPT), and launch tools crafted for specific problems—from legal research at AT&T to sports merch shopping on Fanatics.
It’s not just chatbots answering customer queries anymore. We’re talking about actual work—the kind that makes or breaks productivity. OpenAI’s dataset on workplace ChatGPT use pulls back the curtain for the first time, revealing where generative text is replacing, augmenting, or simply accelerating tasks across continents and industries. Researchers will have a field day, but we should all be asking: Which of these tasks might my own job lose to AI? Is it time to experiment—or to upskill?
The enterprise stories are telling too. “Forward deployed engineers” whose purpose is to help big organizations wring actual value out of increasingly powerful models? That’s not science fiction—it’s today. But notice the subtext: as models get smarter, humans have to get more creative and flexible at integrating them. No wonder infrastructure stories like OpenAI’s Habitat matter. At this scale, split-second improvements can make the difference between an AI tool that’s magical and one that’s forgotten.
We should not, however, treat security and global partnerships as PR fluff. When OpenAI locks down the training of new “frontier” models or partners with international governments, it matters. Security breaches could be catastrophic at these AI scales. And as AI risks become ever more global, collaboration beats going it alone—no matter how fierce the rivalry.
So here’s the challenge for us as users, not just spectators: Are we using these tools only when forced, or are we proactively testing the boundaries? There’s a world of new productivity up for grabs—and this week, it got easier to see (and measure) what’s changing. Try something new at work with AI, or dig into OpenAI’s data. Either way, don’t just read about the future. Shape it.
MORE TOP STORIES
Uber Rolls Out ChatGPT Work Tool to 30,000 Employees Worldwide
- Uber deployed ChatGPT Work (a custom enterprise AI assistant) company-wide, expanding from an initial pilot of 1,000 to over 30,000 employees.
- The tool helps with research, summarization, and other tasks across departments, aiming for a productivity boost.
- This is one of the largest single-organization AI rollouts to date—and a sign of how fast AI is scaling at work.
- Source
AT&T Adopts AI to Reinvent In-House Legal Research
- AT&T’s legal team is expanding its use of OpenAI to organize case facts, documents, and past work—speeding up the process of advising on business decisions.
- The LegalEdge workflow uses AI to surface relevant expertise from across a large, distributed legal department.
- This is among the first detailed reports on AI entering high-stakes corporate legal work at scale.
- Source
OpenAI’s Latest Models Will Soon Be Available on Google Cloud
- OpenAI and Google Cloud are partnering to make ChatGPT and Codex models natively available in Google Cloud’s “Model Garden.”
- The move gives Google Cloud users more options (not just Google’s Gemini models) and lets them build advanced AI agent workflows within Google’s infrastructure.
- Details about pricing and availability are still emerging; it signals more cloud flexibility for AI projects.
- Source
AI Agents Helped Patch Critical Web Vulnerabilities in PHP
- OpenAI’s GPT-5.6 Sol and security partners identified and helped fix severe security gaps in PHP, which powers countless websites.
- The Patch the Planet effort brought together researchers and companies to coordinate rapid discovery and response in recent weeks.
- This demonstrates how AI is now a frontline player in global cybersecurity, not just a background tool.
- Source
ChatGPT Ads Expand to 26 More Countries, Reaching 100,000+ Advertisers
- ChatGPT Ads, which let businesses reach people as they browse and compare products in ChatGPT, are now available in 26 new countries.
- Over 100,000 advertisers have used the platform, with ad options spanning everything from local businesses to global brands.
- As more than 1.2 billion people use ChatGPT each week, AI-powered ads are going global (whether you like it or not).
- Source
Thousands of AI Agents, Millions of Brain Cells: New Research Discovery Workflow
- OpenAI, PerturbAI, and Scripps Research used swarms of AI agents to surface promising leads in gigantic biological datasets, including potential research avenues for diseases like Alzheimer’s.
- AI can now help scientists spot useful patterns in minutes, rather than the months old methods required.
- This signals a dramatic speed-up—and scale-up—in scientific research, with AIs as new lab partners.
- Source
ALSO THIS WEEK
- Cloudflare uses OpenAI to detect and fix security threats in real time — OpenAI Daybreak powers live vulnerability detection and proposed fixes for Cloudflare customers worldwide. (Source)
- Behind ChatGPT’s new ad system — OpenAI shares how it built a novel infrastructure for serving relevant, trusted ads inside ChatGPT. (Source)
- AI halves bookkeeping time for a restaurant franchisee — An independent Schlotzsky’s and Cinnabon owner automated cash reconciliation and freed up $100k in cash flow with ChatGPT Work. (Source)
- GPT-6.1 agent matches senior investor judgment at Terminal X — AI grading system caught 25% more deviations from investment mandates, achieving 88.6% agreement with expert humans. (Source)
- OpenAI acquires Orchestra Bio, an AI-driven drug development firm — The move aims to accelerate drug discovery and bring new treatments to market faster. (Source)
- Find out if your AI model runs on AWS Trainium — Hugging Face adds tool to discover which models will work optimally on Amazon’s specialized AI chips. (Source)
- Database grading becomes RL rewards with Microsoft’s “ThinkingBox” — New method turns disagreements between agents and databases into rewards for better training. (Source)
- NVIDIA Kumo Relational beats traditional ML without needing extra training — New pre-trained model outperformed supervised learning on relational data right out of the box. (Source)
- Meet the AI agents trying to become real journalists — Wired reports on experimental AI agents pitching, researching, and writing news stories on their own. (Source)
- PHP pushes emergency security fix after landmark AI-driven bug discoveries — PHP projects urged to upgrade after major vulnerabilities found by humans and AI agents. (Source)
- Kimi K3.1 Pro language model announced, weights coming soon — New Chinese model available via API now, with open weights promised by November 4. (Source)
- Anthropic donates “Agent Skills” to open AI agent foundation — The move is aimed at keeping agent tech open, community-driven, and not vendor-locked. (Source)
- Software builder’s lament: “AI is our punishment for 30 years of bad code” — Longtime developer Armin Ronacher sums up anxieties about the rapid AI takeover. (Source)
- Audio-first AI real-time strategy game “Battle Drones” debuts — A new voice-controlled, AI-powered RTS lets players battle with drones in a Command & Conquer-style world. (Source)
- “My obituary for programming”: Coding agents change the game — A developer reflects on the career and existential impact of coding agents for programmers. (Source)
- OpenAI’s latest models to appear on Google Cloud and Antigravity — Soon, users can access both OpenAI and Google’s AI models in one place for the first time. (Source)
- FBI finds “Saudi Gov Docs” on accused OpenAI spy’s laptop — Ongoing OpenAI espionage case continues to unfold in U.S. court. (Source)
- The hidden cost of AI: storage, not just GPUs — MIT Technology Review explores how storing AI inference data adds up to unexpected expenses. (Source)
- New Hugging Face AI model lets you “chat” with DNA — The open model decodes and teaches about genetic info, even running on a laptop. (Source)
Want this in your inbox every Monday?
Talk to AI Tech Helper