{
  "docId": "019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "docSlug": "5df6bd1b0447b5f6",
  "documentTitle": "2025 Air Street Capital The State of AI Report 2025",
  "authorId": "AirStreetCapital",
  "authorName": "Air Street Capital",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "vc_research",
  "sourceTypeLabel": "VC research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.777,
  "pageNumber": 82,
  "pageCount": 313,
  "prevPage": 81,
  "nextPage": 83,
  "slideType": "industry_trends",
  "function": "summarize",
  "density": "overcrowded",
  "nDataPoints": 75,
  "notes": "The slide includes a detailed performance table comparing various agents across 15 games.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "comparison_table"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0d0f98caffe1/82",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0d0f98caffe1#slide-82",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "The system sets SOTA across GUI agent benchmarks: 47.5% OSWorld, 50.6% WindowsAgentArena, 73.3% AndroidWorld, 88.2% Online-Mind2Web, and a 59.8 mean normalized score on 15 web games (~60% of human).",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a822-d1e2553187a4",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.6,
        "x": 0.05,
        "y": 0.52
      },
      "kind": "list",
      "text": "The system sets SOTA across GUI agent benchmarks: 47.5% OSWorld, 50.6% WindowsAgentArena, 73.3% AndroidWorld, 88.2% Online-Mind2Web, and a 59.8 mean normalized score on 15 web games (~60% of human), beating OpenAI CUA and Claude Computer Use by large margins.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "6e6ad89e-5047-4719-8c6d-f8080899803d",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.6,
        "x": 0.05,
        "y": 0.35
      },
      "kind": "list",
      "text": "ByteDance's UI-TARS-2 is a native GUI agent trained by collecting trajectories, running supervised fine-tuning followed by multi-turn RL in an all-in-one sandbox with async rollouts and the ability to merge specialized agents.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "c701afc5-e67d-4d4e-9266-d5c3f49f3a88",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.6,
        "x": 0.05,
        "y": 0.8
      },
      "kind": "list",
      "text": "But long-horizon problems remain brittle (e.g., Tetris/Sokoban and hard BrowseComp tasks), and average game skill is ~40% shy of human.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "d826ea27-53b1-4990-b83b-70ed08aca532",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Mean Normalized Score: 59.8",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-41ac-7180-a822-d64b9bb05bfc",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.1,
        "w": 0.9,
        "x": 0.05,
        "y": 0.22
      },
      "kind": "paragraph",
      "text": "Research labs like OpenAI, Anthropic, and ByteDance have been hard at work creating benchmarks and interaction methods for native language model computer use. While the use of RL and multi step reasoning has led to large improvements by and large the models still fall short.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "e60b2880-431b-4871-98a3-388aab1a39cf",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.4,
        "w": 0.3,
        "x": 0.68,
        "y": 0.4
      },
      "kind": "table",
      "text": "Performance comparison table of agents across 15 games.",
      "attrs": null,
      "subkind": "data",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "c3771c59-bf43-4ec5-b667-2ca3dcfdd6ae",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.8,
        "x": 0.05,
        "y": 0.14
      },
      "kind": "title",
      "text": "Computer Use Agents (CUA) have improved by leaps and bounds, and still fall short",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "30a72c55-1caf-48fd-aead-5ea4dcd9dfef",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 189,
      "from": 12,
      "beatId": "019dd95a-0682-776c-8e35-9f380673831f",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Sections 1-2 lay out research findings and industry data with charts and case studies.",
      "position": 1,
      "confidence": 70,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 89,
      "from": 12,
      "beatId": "019dd95a-0682-776c-8e35-ac68557d958b",
      "arcName": "The Onion",
      "arcSlug": "onion",
      "beatName": "First Layer",
      "beatSlug": "onion-first-layer",
      "evidence": "Research section unpacks technical layer beneath the headlines.",
      "position": 2,
      "confidence": 45,
      "parentBeatName": "Development",
      "parentBeatSlug": "development"
    }
  ],
  "loops": [],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}