0:00 Weekly use of Claude inside Microsoft 365 has stalled at less than one-tenth of Copilot's numbers.0:07 In yesterday's episode, we broke down Anthropic's 26 billion dollar debut and whether enterprise adoption would match the paper valuation.0:16 Now, Microsoft's internal data provides the first hard answer.0:21 Across the rest of the landscape, several updates landed overnight.0:26 Alibaba rolled out its flagship Qwen4-Max model, packing two point four trillion parameters and claiming coding wins over GPT-6.1-Sol.0:35 In India, Sarvam AI launched Token Factory, a localized inference stack designed to run open-weight models cheaper on domestic GPU clusters.0:45 Zoho introduced Blended Agents for Zoho Desk, allowing human supervisors to halt autonomous customer actions mid-stream.0:54 Arborum Labs secured pre-seed funding to build infrastructure world models.0:59 And Nvidia's board officially ruled out the return of Jensen Huang after his resignation, focusing entirely on finding his successor.1:09 Look, the gap between model capability and enterprise workflow is where AI narratives go to die.1:15 Anthropic launched Claude integrations for Word, Excel, and PowerPoint in May 2026.1:21 On paper, Claude remains a researcher favorite for deep analysis.1:26 But inside enterprise software, convenience eats capability every single day.1:32 Copilot sits on the default ribbon, backed by corporate procurement and administrative IT sign-off.1:39 When users have to step outside their pre-configured lane to invoke an external agent, ninety percent of them simply don't bother.1:48 Distribution didn't just protect Microsoft's moat.1:51 It neutralized an architectural rival without firing a shot.1:56 Okay, let's turn to the other major data point today: research acceleration.2:01 OpenAI published an analysis of more than five hundred and eighty thousand scientific exchanges across eighteen hundred researchers.2:11 The headline figure is striking: ninety-eight percent of surveyed scientists expect artificial intelligence to accelerate discoveries, and over one in five believe a decade-long breakthrough can be compressed into three years.2:27 Well, don't mistake researcher optimism for bench-science reality.2:31 Here's the thing: generating hypotheses and drafting literature reviews takes half the time, but physical verification still runs on biological clocks and wet-lab throughput.2:44 OpenAI's own report admits that funding, personnel, and physical constraints remain the actual bottlenecks to scientific output.2:53 Software can propose a molecule in forty-eight hours, yet proving it works in a living organism still takes years of clinical grit.3:02 So, we're left with two distinct realities this week.3:06 The benchmark race continues to produce dizzying numbers, but in the enterprise and in the lab, raw compute hits the friction of human institutions.3:17 Anyway, access to the prompt means nothing if your tool can't survive the friction of the enterprise default.