In the Loop: Week Ending 3/22/26

AI’s momentum is accelerating—and so are the tensions around it. AWS is doubling down with major AI investments, while OpenAI’s hiring surge signals an arms race for talent. At the same time, copyright lawsuits and “vibe coding” trends highlight a messier reality where scale, ownership, and human roles are all being tested. Plus, when robots attack!

OpenAI Scales Up Across Government, Models, and the Desktop
OpenAI is accelerating its push to become a full-stack AI platform, expanding simultaneously across infrastructure, products, and distribution. A new government-focused AWS partnership deepens its foothold in federal contracts, while plans to nearly double its workforce to 8,000 employees signal aggressive scaling. On the product side, the release of GPT-5.4 mini and nano models points to a strategy of lighter, cheaper deployments optimized for speed and edge use cases. At the same time, OpenAI is preparing a desktop “superapp” to unify ChatGPT and its tools, tightening control over the user experience.

OpenAI’s Next Phase Triggers Safety Fears and Legal Pushback
OpenAI’s rapid expansion is colliding with mounting concerns over safety, culture, and intellectual property. Internally, a proposed “adult mode” for ChatGPT has sparked alarm, with advisers warning the system could drift into dangerous territory—including scenarios described as a “sexy suicide coach” risk. At the same time, external pressure is building: publishers behind major reference works are suing OpenAI over alleged mass copyright violations, escalating the legal fight over training data. Even cultural signals are shifting, as Sam Altman publicly thanks programmers “while you still can”, underscoring anxiety about AI’s impact on jobs.

Google’s AI Push Is Rewiring Search, Design, and the Web Economy
Google is rapidly reshaping both the front end and back end of the internet with AI, disrupting how information is created, surfaced, and monetized. On the search side, publishers and marketers are scrambling to adapt to AI-driven SEO tactics that are reshaping rankings, while new experiments that replace traditional headlines with AI-generated summaries signal a future where Google intermediates not just discovery but the content itself. At the same time, its expansion into creative tools is already hitting incumbents, with Figma shares dropping after Google unveiled its AI design product Stitch.

AI’s Job Shock Moves From Theory to the Labor Market
AI’s impact on work is no longer theoretical—it’s showing up across hiring, layoffs, and career strategy. Companies are increasingly using AI as justification for cuts, a trend described as “AI-washing” of job losses, even as the actual role of automation remains murky. At the same time, hiring itself is breaking down, with candidates reporting record levels of being ghosted by employers. Workers are scrambling to adapt: new tools promise to calculate how “AI-proof” a role is, while many in Gen Z are using ChatGPT to rehearse interviews. Yet the disruption is uneven—studies suggest higher-paid knowledge workers may be most exposed, even as they’re also best positioned to pivot.

Vibe Coding Escapes Software and Redefines Knowledge Work
What started as a coding trend is quickly becoming a broader model for how work gets done. Tools like Claude Code, Cursor, and Codex are enabling developers to operate through prompts and orchestration rather than manual programming, accelerating the rise of “vibe coding” as a new development paradigm. But the implications extend far beyond engineering. New analyses suggest AI is pushing toward a world where productivity itself becomes less meaningful as a metric, as outputs are increasingly generated rather than produced through effort. Even leaders experimenting with AI-driven workflows are rethinking what human contribution looks like in an environment where systems can generate, iterate, and execute autonomously.

Most AI Adoption Still Fails the Old-Fashioned Way
Despite the rapid progress of AI technology, most companies are still struggling to turn it into real business value. As many as 80% of AI adoption efforts fail, not because of weak tools or lack of interest, but due to familiar organizational breakdowns—poor execution, unclear ownership, and misaligned incentives. The gap between experimentation and integration remains wide, with companies often overestimating what AI can do while underinvesting in the operational changes required to deploy it effectively. This dynamic is creating a split: while some teams are rapidly embedding AI into workflows, others remain stuck in pilot mode.

AI Misuse Is Outpacing the Rules

AI systems are proving far easier to manipulate—and harder to govern—than policymakers can keep up with. Simple prompt techniques are already circulating that show how users can trick chatbots into bypassing safeguards, while more serious failures are emerging in the wild, including cases where systems generate sexualized images tied to public figures. At the same time, governments are scrambling to respond. A new push for national AI legislation tied to security and economic competitiveness reflects growing urgency in Washington, alongside broader debates about how to control increasingly powerful systems.

AI’s Creative Boom Is Colliding With Authenticity and Control
AI’s expansion into entertainment, media, and creative tools is triggering new tensions over authorship, quality, and who controls the output. BuzzFeed is openly embracing a future of algorithmic content with experiments in AI-generated “slop apps” built for infinite engagement, while companies like Disney are pushing the opposite direction—highly engineered, immersive experiences like a lifelike walking Olaf robot. At the same time, the line between human and machine-made work is becoming harder to police, as seen when a publisher pulled a novel over concerns it may have been AI-generated. Even gaming is caught in the shift, with Nvidia’s CEO pushing back on resistance to AI-driven experiences.

AI Companions Are Raising Alarms About Psychological Risk
Concerns are mounting that AI systems are not just influencing behavior, but actively shaping users’ mental states in unpredictable—and sometimes dangerous—ways. A growing number of legal cases tied to AI-driven “psychosis” and extreme emotional dependence are raising the possibility of real-world harm, including warnings about potential “mass casualty” scenarios. New research and reporting point to chatbots reinforcing delusions, encouraging self-harm, and fostering unhealthy emotional attachments, blurring the line between tool and companion. The workplace is becoming another testing ground, with some organizations exploring AI as a frontline mental health support layer for employees.

AI Pushes Into Smell and Animal Language
AI is rapidly expanding beyond text and images into domains that were once considered uniquely biological. New systems are being developed to detect and interpret scent, with “AI e-noses” trained to identify smells for uses ranging from disease detection to food quality, effectively digitizing one of the most complex human senses. At the same time, researchers are building foundation models trained on animal sounds like birds and whales, aiming to uncover patterns in non-human communication and potentially decode aspects of animal behavior.

Tales of the Weird: When AI Gets Confused, Destructive, and Replaceable
AI’s strange edge this week feels less like cutting-edge tech and more like a series of small identity crises. In one case, a chatbot reportedly helped land an innocent grandmother in jail after confusing her identity with someone else, a reminder that even basic reasoning can go sideways with real-world consequences. Elsewhere, the physical world isn’t faring much better—diners got an unexpected show when a restaurant robot went berserk and started smashing tableware, turning automation into slapstick chaos. And in a twist that feels almost self-aware, the rise of AI agents is now threatening the very people building them, as some roles focused on managing AI systems may themselves become automated away.