{
  "docId": "019dd923-5e88-73ef-bd5d-47815abb602b",
  "docSlug": "485b1e996a637b43",
  "documentTitle": "2025 AI Agents and the Model Context Protocol",
  "authorId": "BCG",
  "authorName": "BCG",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "strategy_consulting",
  "sourceTypeLabel": "Strategy consulting",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.778,
  "pageNumber": 14,
  "pageCount": 37,
  "prevPage": 13,
  "nextPage": 15,
  "slideType": "industry_trends",
  "function": "summarize",
  "density": "overcrowded",
  "nDataPoints": 1,
  "notes": "The slide highlights the shift in AI benchmarking from simple tasks to complex, multi-turn, and robust agent evaluation.",
  "elementsJson": [
    "bullet_list",
    "screenshot",
    "line_chart",
    "quote_block",
    "icon_grid"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-47815abb602b/14",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-47815abb602b",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-47815abb602b.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-47815abb602b#slide-14",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "RE-Bench Evaluating frontier AI R&D capabilities of language model agents against human experts",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c6-7671-887d-7ec2bde47835",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.28,
        "w": 0.28,
        "x": 0.66,
        "y": 0.51
      },
      "kind": "chart",
      "text": "Best Observed Score@k by Time Budget",
      "attrs": null,
      "subkind": "line",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "fed25427-3f58-4517-9644-5b7f15424742",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.35,
        "w": 0.2,
        "x": 0.51,
        "y": 0.24
      },
      "kind": "image",
      "text": "Columbia Journalism Review article snippet",
      "attrs": null,
      "subkind": "screenshot",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "de317278-1636-4bf0-9276-049629d25041",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.3,
        "x": 0.15,
        "y": 0.25
      },
      "kind": "list",
      "text": "Benchmarks are shifting to measuring how well agents use tools and handle end-to-end tasks over time within different domains",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "1f382b96-5ae4-4f99-b1bc-ac5954c1bfa4",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.3,
        "x": 0.15,
        "y": 0.45
      },
      "kind": "list",
      "text": "New test cases emphasize edge scenarios like missing tools, irrelevant queries, and incomplete inputs to probe agent robustness",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "6c67391e-d0e4-47e8-bfd8-582f78a0079f",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.3,
        "x": 0.15,
        "y": 0.65
      },
      "kind": "list",
      "text": "Increasing focus on multi-turn tasks requires agents to manage context, sequence actions, and adapt to evolving goals (e.g. ML training)",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "c76b6ee9-4817-45fb-8626-b1113de12c9c",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Average Normalized Score",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c6-7671-887d-845219fd3cc6",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "quote",
      "text": "“AI Search Has a Citation Problem: We Compared Eight AI Search Engines. They’re All Bad at Citing News.” Columbia Journalism Review",
      "attrs": null,
      "subkind": null,
      "toolName": "Authority citation",
      "toolSlug": "authority-citation",
      "confidence": null,
      "componentId": "019dd952-47c6-7671-887d-827c7fb2f88c",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.03,
        "w": 0.6,
        "x": 0.05,
        "y": 0.94
      },
      "kind": "source-note",
      "text": "Source: AI Search Has A Citation Problem - Columbia Journalism Review; Galileo AI; Scale AI; BCG analysis; Expert interviews",
      "attrs": null,
      "subkind": null,
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "dbe46997-09ef-44fe-813a-4ab13e442883",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.5,
        "x": 0.05,
        "y": 0.09
      },
      "kind": "title",
      "text": "Benchmarks and studies are exploring agent reliability",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "0ec4f436-f889-45c2-8ff1-d71ef687db53",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Aha! Moment",
      "slug": "aha-moment",
      "agent": "Storyteller",
      "layer": "loop",
      "matchId": "019dd95a-1155-76cc-9e44-9c592c5ac8f4",
      "evidence": "First page of the aha-moment loop sets up the question to be revealed",
      "confidence": 75
    },
    {
      "name": "Action Titles",
      "slug": "action-titles",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "019dd95a-1155-76cc-9e44-94f7ace1e579",
      "evidence": "Title 'Benchmarks and studies are exploring agent reliability'",
      "confidence": 80
    },
    {
      "name": "Authority Bias",
      "slug": "authority-bias",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-1155-76cc-9e44-9bd241ae8cb9",
      "evidence": "References RE-Bench and frontier-AI evaluation literature",
      "confidence": 75
    },
    {
      "name": "List presentation",
      "slug": "list-presentation",
      "agent": null,
      "layer": "slide",
      "matchId": "70bd8477-11bb-4581-8c2d-8148bdb5dba1",
      "evidence": "The slide contains a list/bullet points with text.",
      "confidence": 0.7
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 17,
      "from": 10,
      "beatId": "019dd95a-0682-776c-8e39-efdc0c3bfe92",
      "arcName": "The Consultant's Gambit",
      "arcSlug": "consultants-gambit",
      "beatName": "Problem & Complication",
      "beatSlug": "consultants-gambit-problem-complication",
      "evidence": "PMF reached only narrowly; 6-dim assessment exposes capability gaps",
      "position": 2,
      "confidence": 82,
      "parentBeatName": "Complication",
      "parentBeatSlug": "complication"
    },
    {
      "to": 17,
      "from": 13,
      "beatId": "019dd95a-0682-776c-8e3a-022aa9a7736c",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Implications (So What)",
      "beatSlug": "triple-take-the-implications-so-what",
      "evidence": "Section 03 reliability benchmarks and BCG capability framework",
      "position": 2,
      "confidence": 55,
      "parentBeatName": "Reflection",
      "parentBeatSlug": "reflection"
    }
  ],
  "loops": [
    {
      "to": 15,
      "from": 14,
      "name": "Aha Moment",
      "slug": "03-aha-moment",
      "bestFor": "Data-heavy sections, research findings, analytical arguments",
      "matchId": "019dd95a-088b-72c8-b7df-f64c44f43b3b",
      "evidence": "p14 sets up agent benchmarks; p15 reveals METR exponential with 'month-long projects' dark callout",
      "position": 3,
      "objective": "Reveal the 7-month doubling capability insight",
      "structure": "The Problem/Question -> What the Data Shows -> The Insight",
      "confidence": 80,
      "description": "Establish a tension, present data, then deliver the insight that changes everything"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}