The market is consolidating around execution while regulators scramble to slow it down. OpenAI is killing consumer products like Sora to chase enterprise deals and a potential IPO, Microsoft is locking Copilot Chat behind paywalls, and venture capital is flooding into startups that can actually deploy agents to do real work: Harvey at 11 billion dollars, Granola jumping from 250 million to 1.5 billion, Lucid Bots raising capital to keep up with demand for window-washing robots. Meanwhile, the builders are moving faster than the gatekeepers. Anthropic is adding auto mode to Claude Code so agents can execute on user behalf with safeguards, Cloudflare rolled out Dynamic Workers for AI agent execution in milliseconds, and Oracle added prebuilt agents to its database. Google's TurboQuant compresses LLM memory by six times without degrading output, a straightforward efficiency win that matters for deployment at scale. These are not research papers or frameworks. These are products shipping into production.
The friction points reveal where power is actually contested. Reddit is requiring human verification for suspicious accounts while still allowing AI-generated content, a half-measure that protects against bot spam without blocking the commercial use case. Bernie Sanders and Alexandria Ocasio-Cortez introduced legislation to halt data center construction until Congress passes comprehensive AI regulation, a blunt instrument that acknowledges their inability to regulate the technology itself so they are targeting the infrastructure instead. Disney was blindsided by OpenAI's decision to shut down Sora, suggesting the partnership was never real enough to warrant a heads-up before the company pivoted to enterprise. PyPI had to warn developers after malicious versions of LiteLLM stole cloud and CI/CD credentials, exposing how fragile the supply chain is when open-source middleware becomes critical infrastructure for AI applications. The gap between who controls the tools and who controls the rules is widening.
Workforce inequality is emerging as the actual problem, not safety. Anthropic's own data shows AI is not replacing jobs yet, but early adopters with the skills to use these tools are pulling ahead while everyone else falls behind. Deccan AI raised 25 million dollars and concentrated its workforce in India to manage quality in the AI training market, a straightforward arbitrage play on labor costs. Google is launching Lyria 3 Pro for music generation, Meta is using generative AI for shopping recommendations, HP is cramming a 20-billion-parameter model into enterprise laptops, and Axiom Math released Axplorer free to mathematicians to discover patterns. The winners are companies that can move product into users' hands and extract value from their behavior. The losers are people waiting for regulation that will never move as fast as deployment.
Sloane Duvall