The real tension in AI deployment today isn't between skeptics and believers. It's between vendors extracting value through opacity and enterprises trying to build stable systems on shifting sand. A Harvard study showing LLMs outperforming human doctors in emergency room diagnosis sounds like vindication for AI adoption, but it arrives in a week when enterprise IT leaders are quietly panicking about vendors unilaterally degrading the systems they've already paid billions for, with no notification and no recourse. Artisan's billboard campaign urging companies to stop hiring humans uses stolen art from a creator, which is less a scandal than a perfect illustration of the incentive structure: move fast, grab whatever works, apologize later if caught. The pattern repeats across sectors. Water utilities in Singapore cut leakage to 75% below England and Wales using AI. Restaurants deploy the technology to reduce waste. Recruiters use it to clear administrative work so humans can focus on judgment calls. Hedge funds and wealth managers are adopting AI for speed and document analysis while keeping it away from truly sensitive decisions. These aren't revolutionary claims. They're ordinary businesses using a tool where it reduces cost or error in bounded, measurable ways.
The infrastructure underneath is where the real leverage sits. Mistral AI's new 128B model scores 77.6% on SWE-Bench Verified, pushing the frontier on coding agents. Sakana AI's KAME injects LLM knowledge into speech-to-speech systems without adding latency. Developers are learning to systematize prompting, moving beyond trial-and-error iteration toward reliability engineering. These are the moves that matter: not grand claims about transformation, but the unglamorous work of making production systems stable enough that enterprises can actually depend on them. Yet the Computerworld report on vendor lock-in cuts deeper. When LLMs change behavior without warning, when tokenization drift can degrade a model's performance without any change to data or logic, when IT has little visibility into what's happening inside systems they control, enterprises face a choice between fragility and capitulation. The vendors aren't necessarily acting in bad faith. But they've structured the relationship so that enterprises absorb all the operational risk while vendors retain all the control. That's the real story beneath today's headlines.
Sloane Duvall