{
 "title": "Today's AI News: The Big Picture",
 "description": "Today's AI News: The Big Picture\nToday's essential takeaway: Spending intensity shift—the Ramp Index is now tracking AI spending per employee instead of adoption rates.\n\nKey segments in this episode\n- Today's Big Picture\n- Ramp Index switches to spending intensity\n- Anthropic takes the #1 corporate adoption spot\n- Next-gen models: GPT-5.6 and Gemini 3.5\n- Regulatory timeline: August and December\n\nDrop a comment with the topic from today's news you'd like us to dig deeper into.\nIf you see the Hype button in the last 7 days of uploads, give it a tap—it helps new viewers find this show.\n\nThis broadcast is brought to you by lawyer Kyungjin Kim.\n\n#AI #ArtificialIntelligence #KyungjinKim",
 "link": "https://kimkj.com/%EC%98%A4%EB%8A%98%EC%9E%90-%EB%89%B4%EC%8A%A4-%EC%98%81%EC%83%81/?mod=document&uid=5906",
 "segments": [
  {
   "id": "S001",
   "slide": 1,
   "speaker": "Host",
   "text": "This broadcast is brought to you by lawyer Kyungjin Kim."
  },
  {
   "id": "S002",
   "slide": 1,
   "speaker": "Guest",
   "text": "What's the number you need to look at first today?"
  },
  {
   "id": "S003",
   "slide": 1,
   "speaker": "Host",
   "text": "Ramp looked at actual payment data from about 70,000 companies. The top 1 percent of businesses? They're spending about 7,500 dollars per employee per month on AI. The median is 11.38 dollars. The gap in how deep companies are going is so wide that \"same company\" starts to sound meaningless. It signals the end of the era when companies were asking whether to adopt AI or not."
  },
  {
   "id": "S004",
   "slide": 1,
   "speaker": "Guest",
   "text": "Has that shift shaken up the rankings?"
  },
  {
   "id": "S005",
   "slide": 1,
   "speaker": "Host",
   "text": "Yeah. Anthropic just overtook OpenAI in paid adoption rates for the first time, and they're pulling further ahead. Meanwhile, SpaceX-xAI hit the market with the biggest IPO ever. Today we're focusing on these two moments—plus we're going to dig into regulation and the supply chain."
  },
  {
   "id": "S006",
   "slide": 2,
   "speaker": "Guest",
   "text": "Why'd they start tracking spending instead of adoption rates?"
  },
  {
   "id": "S007",
   "slide": 2,
   "speaker": "Host",
   "text": "Because adoption rates were getting close to 100 percent—they stopped being useful as a differentiator. Once everyone's using something, what actually matters is who's using it deeper. That's where the real difference comes from. Ramp switched their metric to monthly spending per employee. If one company's at 7,500 and the next one's at 11 dollars, you can't really say they're using the same tool."
  },
  {
   "id": "S008",
   "slide": 2,
   "speaker": "Guest",
   "text": "How should practitioners take this number?"
  },
  {
   "id": "S009",
   "slide": 2,
   "speaker": "Host",
   "text": "I'd recommend calculating where your company sits on per-person AI spending. If you don't know your position, you can't gauge how far competitors have lifted their workflow speeds—and you might fall behind without realizing it. You also need to see that the top tier is rotating through multiple models."
  },
  {
   "id": "S010",
   "slide": 3,
   "speaker": "Host",
   "text": "This broadcast is brought to you by lawyer Kyungjin Kim."
  },
  {
   "id": "S011",
   "slide": 3,
   "speaker": "Guest",
   "text": "Where's Anthropic's edge coming from?"
  },
  {
   "id": "S012",
   "slide": 3,
   "speaker": "Host",
   "text": "It comes from tools that snap right into real workflows. Just Claude Code alone—the coding assistant—pulled in a billion dollars in annual revenue. It's not about feature demos. It's that it actually cuts developers' daily workload, so the payments followed. The share of companies picking Claude as their first choice when they're new to AI has gone up too."
  },
  {
   "id": "S013",
   "slide": 3,
   "speaker": "Guest",
   "text": "So there are no weaknesses?"
  },
  {
   "id": "S014",
   "slide": 3,
   "speaker": "Host",
   "text": "Yeah, there are. As Anthropic builds out its own vertical products, it's starting to collide with partners who were using its models. During the same period, Figma's stock—the design collaboration tool—got cut in half. It's a perfect picture of what happens to partnerships when the company that was powering you becomes the competition."
  },
  {
   "id": "S015",
   "slide": 4,
   "speaker": "Guest",
   "text": "There's no official announcement yet, so why cover it now?"
  },
  {
   "id": "S016",
   "slide": 4,
   "speaker": "Host",
   "text": "Context window is the amount of text a model can read at once. At the rumored 1.5 million tokens, you could dump a whole stack of thick contracts in and review them all at once. Gemini 3.5 is supposed to hit 2 million tokens. When the volume of documents you can handle at once goes up, how you design work itself changes. That said, both specs are pre-launch observations, so nothing's confirmed yet."
  },
  {
   "id": "S017",
   "slide": 4,
   "speaker": "Guest",
   "text": "What's the change users need to handle right now?"
  },
  {
   "id": "S018",
   "slide": 4,
   "speaker": "Host",
   "text": "The GPT-5.2 model family retired on June 12th, and existing conversations are auto-converting to GPT-5.5. If you've been running automation scripts with a specific model number, it's safer to check the calling names. If you don't notice the model switched, your next run hits an error or gets a different model responding."
  },
  {
   "id": "S019",
   "slide": 5,
   "speaker": "Guest",
   "text": "What changes in August?"
  },
  {
   "id": "S020",
   "slide": 5,
   "speaker": "Host",
   "text": "The EU AI Act's mandatory provisions for high-risk fields kick in. If you're using AI in places that directly touch human rights—hiring, credit scoring, law enforcement—you've got obligations around documentation, transparency, and human oversight. Break them, you can face fines up to 3 percent of global revenue. Korean companies serving Europe are on the hook too."
  },
  {
   "id": "S021",
   "slide": 5,
   "speaker": "Guest",
   "text": "What's the New York side like?"
  },
  {
   "id": "S022",
   "slide": 5,
   "speaker": "Host",
   "text": "A law banning AI companion chatbots that mimic humans for anyone under 18 passed unanimously. No split vote means child protection had almost no political pushback. The governor has until December 31st to sign. If you're running services aimed at minors, you need to check your age verification and chatbot personality design now."
  },
  {
   "id": "S023",
   "slide": 6,
   "speaker": "Host",
   "text": "This broadcast is brought to you by lawyer Kyungjin Kim."
  },
  {
   "id": "S024",
   "slide": 6,
   "speaker": "Guest",
   "text": "Why does the TCS deal matter?"
  },
  {
   "id": "S025",
   "slide": 6,
   "speaker": "Host",
   "text": "In heavily regulated fields like finance, government, and healthcare, AI gets stuck in the pilot phase. Data protection and liability are the bottlenecks. TCS has the connections to get into those industries, and they're saying they'll roll Claude out to 50,000 employees and build joint solutions. They're aiming to be the bridge that moves things past pilots into actual production work."
  },
  {
   "id": "S026",
   "slide": 6,
   "speaker": "Guest",
   "text": "How far along is Japan's surgical AI?"
  },
  {
   "id": "S027",
   "slide": 6,
   "speaker": "Host",
   "text": "It reads organ and blood vessel images, then generates step-by-step surgical guidance. Expert review showed 85 to 90 percent accuracy. Japan—which faces a serious surgeon shortage—is pushing this with government backing. It's not in clinical use yet, so they're still setting evaluation standards. In medicine, where mistakes cost lives, you can't decide to deploy based on accuracy numbers alone. That's the bigger picture."
  },
  {
   "id": "S028",
   "slide": 7,
   "speaker": "Guest",
   "text": "You mentioned the biggest IPO ever—how big are we talking?"
  },
  {
   "id": "S029",
   "slide": 7,
   "speaker": "Host",
   "text": "SpaceX-xAI raised 75 billion dollars all at once, and the market cap jumped to around 1.77 trillion. Alphabet's 4.9 percent stake alone is worth about 105 billion dollars. On the second trading day, it was still holding around 161 dollars. The market is putting a premium on the structure of bundling space and AI under one company."
  },
  {
   "id": "S030",
   "slide": 7,
   "speaker": "Guest",
   "text": "How do Anthropic and OpenAI's finances stack up differently?"
  },
  {
   "id": "S031",
   "slide": 7,
   "speaker": "Host",
   "text": "Anthropic hit 10.9 billion in quarterly revenue with 559 million in operating profit—they're eyeing their first profitable quarter since launch. But they've already flagged that infrastructure spending means they won't hold onto that profit starting next quarter. OpenAI's annualized revenue topped 25 billion, but one analyst's math says they spend 1.22 dollars for every dollar they earn. Two companies that both filed confidential S-1s with nearly opposite profitability structures—investors need to see that distinction."
  },
  {
   "id": "S032",
   "slide": 8,
   "speaker": "Guest",
   "text": "With China's investment, is there something bigger than the dollar amount?"
  },
  {
   "id": "S033",
   "slide": 8,
   "speaker": "Host",
   "text": "The conditions are the key part. They're putting in 295 billion dollars, but there's a string attached: at least 80 percent of components have to be from China. Coupled with U.S. export controls, you're watching the AI infrastructure supply chain split into a U.S. bloc and a China bloc. Same AI, but which chip it runs on and which model it sits on top of—that varies by country now."
  },
  {
   "id": "S034",
   "slide": 8,
   "speaker": "Guest",
   "text": "Pull together what practitioners need to check first thing tomorrow morning."
  },
  {
   "id": "S035",
   "slide": 8,
   "speaker": "Host",
   "text": "Check three things. Pull the number on your company's monthly per-person AI spending and gauge your competitive position. Check whether the model names baked into your automation scripts are retired versions like GPT-5.2. If you're serving Europe-based customers or running services for minors, sync with legal to see how the August 2 EU high-risk obligations and New York's child chatbot law apply to you."
  },
  {
   "id": "S036",
   "slide": 8,
   "speaker": "Host",
   "text": "This broadcast is brought to you by lawyer Kyungjin Kim."
  }
 ]
}