OpenAI and Anthropic launched fresh products this week that expose starkly different bets on how AI becomes indispensable. OpenAI is building an app ecosystem atop ChatGPT itself—Prism for academic writing, Sora for images, a health tool, Code Interpreter—each a discrete interface within a single platform. It's the app-store-as-operating-system play. Anthropic, by contrast, is pushing apps deeper into its desktop Claude experience, prioritizing a more integrated, less fragmented approach. The distinction matters. OpenAI's strategy assumes ChatGPT becomes the infrastructure; Anthropic's suggests Claude remains the product. Neither has released a meaningfully new foundational model in months, which signals a shift: the industry is no longer chasing raw capability increases. William Sutskever, a co-founder of OpenAI's safer-AI division who now runs Safe Super Intelligence, has called this "back to the age of research"—a reminder that the venture-backed AI company model may be broken. If scaling neural networks has hit a wall, pouring billions into inference compute becomes a diminishing return. The real money, he implies, goes to unfunded research labs willing to chase the next architectural paradigm shift, not to teams productizing GPT-5.
The funding battles tell the story better than the products. OpenAI is raising $100 billion to reach an $830 billion valuation; Anthropic is targeting $20–30 billion for a $350 billion valuation. On paper, the gap is staggering. In reality, neither company is close to profitable. OpenAI's Amazon deal ($50 billion) and SoftBank injection ($30 billion) are largely circular: the capital funds compute contracts with—Amazon. It's a clever closed loop, but it relies on Sam Altman's singular ability to extract capital from a dwindling pool of mega-fund dry powder. Anthropic's approach appears more constrained, suggesting an IPO is likelier within the next two years. When either company goes public, the liquidity event could reshape venture capital itself, flooding the market with founders' capital and accelerating the shift toward unfunded, high-risk research bets.
What actually matters for employment, though, is not what OpenAI and Anthropic build next, but when their tools become good enough to be invisible. An Economist analysis of US labour data since 1982 found no collapse in white-collar employment—software developer roles grew 7% over thirteen years, radiologist positions up 10%, paralegals up 21%. Even since ChatGPT's launch, the trend continues upward. The study's claim: AI will redistribute work, not eliminate it. Code assistants are the early proof point. They've only become genuinely useful in the past six months, yet adoption is already reshaping the role. A developer now spends less time writing boilerplate and more time on architecture and judgment calls. Tomorrow's developer does less, but is expected to do more—a classic productivity paradox that repeats through every wave of automation. The risk is narrower but real: workers whose skills are non-transferable—those whose entire value is a single, automatable function—face genuine disruption. A software developer who can't think beyond syntax is more vulnerable than one who understands systems design. But those workers have always been vulnerable. What changes is the timeline.