🐦 Twitter Post Details

Viewing enriched Twitter post

@omarsar0

Cool paper from Apple. Most evaluation of tool-calling agents happens after the trajectory is over. By then the wrong call has already shipped. This new paper moves evaluation into the execution loop. A specialized reviewer agent inspects each provisional tool call before it executes. If something is off, it injects feedback and the primary agent revises. To quantify the tradeoff between corrections and new mistakes, they introduce Helpfulness-Harmfulness metrics. Helpfulness measures the percentage of base errors fixed; harmfulness measures correct calls degraded by the reviewer. Results on BFCL: +5.5% on irrelevance detection (84.9% to 90.4%), +1.6% on relevance, all with no retraining of the base agent. On τ²-Bench multi-turn: +7.1% (48.7% to 55.8%). Reasoning-model reviewers get a 3:1 benefit-to-risk ratio vs. 2.1:1 for GPT-4o. Adding GEPA prompt optimization stacks another +1.5–2.8%. Why does it matter? You can keep the base tool-calling agent frozen and still ship measurable accuracy gains by improving only the reviewer. Model selection and prompt optimization on the reviewer become real, separable production levers. Paper: https://t.co/L0p0UBFcI0 Learn to build effective AI agents in our academy: https://t.co/1e8RZKs4uX

Media 1

📊 Media Metadata

{
  "media": [
    {
      "type": "photo",
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2053559715289047280/media_0.jpg",
      "filename": "media_0.jpg"
    }
  ],
  "processed_at": "2026-05-11T22:29:10.213605",
  "pipeline_version": "2.0"
}

🔧 Raw API Response

{
  "type": "tweet",
  "id": "2053559715289047280",
  "url": "https://x.com/omarsar0/status/2053559715289047280",
  "twitterUrl": "https://twitter.com/omarsar0/status/2053559715289047280",
  "text": "Cool paper from Apple.\n\nMost evaluation of tool-calling agents happens after the trajectory is over. By then the wrong call has already shipped.\n\nThis new paper moves evaluation into the execution loop. A specialized reviewer agent inspects each provisional tool call before it executes. If something is off, it injects feedback and the primary agent revises.\n\nTo quantify the tradeoff between corrections and new mistakes, they introduce Helpfulness-Harmfulness metrics. Helpfulness measures the percentage of base errors fixed; harmfulness measures correct calls degraded by the reviewer.\n\nResults on BFCL: +5.5% on irrelevance detection (84.9% to 90.4%), +1.6% on relevance, all with no retraining of the base agent.\n\nOn τ²-Bench multi-turn: +7.1% (48.7% to 55.8%).\n\nReasoning-model reviewers get a 3:1 benefit-to-risk ratio vs. 2.1:1 for GPT-4o. Adding GEPA prompt optimization stacks another +1.5–2.8%.\n\nWhy does it matter?\n\nYou can keep the base tool-calling agent frozen and still ship measurable accuracy gains by improving only the reviewer. Model selection and prompt optimization on the reviewer become real, separable production levers.\n\nPaper: https://t.co/L0p0UBFcI0\n\nLearn to build effective AI agents in our academy: https://t.co/1e8RZKs4uX",
  "source": "Twitter for iPhone",
  "retweetCount": 41,
  "replyCount": 31,
  "likeCount": 260,
  "quoteCount": 1,
  "viewCount": 15557,
  "createdAt": "Sun May 10 19:36:04 +0000 2026",
  "lang": "en",
  "bookmarkCount": 257,
  "isReply": false,
  "inReplyToId": null,
  "conversationId": "2053559715289047280",
  "displayTextRange": [
    0,
    277
  ],
  "inReplyToUserId": null,
  "inReplyToUsername": null,
  "author": {
    "type": "user",
    "userName": "omarsar0",
    "url": "https://x.com/omarsar0",
    "twitterUrl": "https://twitter.com/omarsar0",
    "id": "3448284313",
    "name": "elvis",
    "isVerified": false,
    "isBlueVerified": true,
    "verifiedType": null,
    "profilePicture": "https://pbs.twimg.com/profile_images/939313677647282181/vZjFWtAn_normal.jpg",
    "coverPicture": "https://pbs.twimg.com/profile_banners/3448284313/1565974901",
    "description": "",
    "location": "DAIR.AI Academy",
    "followers": 303366,
    "following": 830,
    "status": "",
    "canDm": true,
    "canMediaTag": true,
    "createdAt": "Fri Sep 04 12:59:26 +0000 2015",
    "entities": {
      "description": {
        "urls": []
      },
      "url": {}
    },
    "fastFollowersCount": 0,
    "favouritesCount": 35835,
    "hasCustomTimelines": true,
    "isTranslator": true,
    "mediaCount": 4645,
    "statusesCount": 17833,
    "withheldInCountries": [],
    "affiliatesHighlightedLabel": {},
    "possiblySensitive": false,
    "pinnedTweetIds": [
      "2053575378208256086"
    ],
    "profile_bio": {
      "description": "Building @dair_ai • Prev: Meta AI, Elastic, PhD • New AI learning portal: https://t.co/1e8RZKs4uX",
      "entities": {
        "description": {
          "urls": [
            {
              "display_url": "academy.dair.ai",
              "expanded_url": "https://academy.dair.ai/",
              "indices": [
                74,
                97
              ],
              "url": "https://t.co/1e8RZKs4uX"
            }
          ],
          "user_mentions": [
            {
              "id_str": "",
              "indices": [
                9,
                17
              ],
              "name": "",
              "screen_name": "dair_ai"
            }
          ]
        },
        "url": {
          "urls": [
            {
              "display_url": "dair.ai",
              "expanded_url": "https://www.dair.ai/",
              "indices": [
                0,
                23
              ],
              "url": "https://t.co/XQto5ypSIk"
            }
          ]
        }
      }
    },
    "isAutomated": false,
    "automatedBy": null
  },
  "extendedEntities": {
    "media": [
      {
        "display_url": "pic.twitter.com/JKCt67mfWJ",
        "expanded_url": "https://twitter.com/omarsar0/status/2053559715289047280/photo/1",
        "ext_media_availability": {
          "status": "Available"
        },
        "features": {
          "large": {
            "faces": []
          },
          "orig": {
            "faces": []
          }
        },
        "id_str": "2053559711631585280",
        "indices": [
          278,
          301
        ],
        "media_key": "3_2053559711631585280",
        "media_results": {
          "id": "QXBpTWVkaWFSZXN1bHRzOgwAAQoAARx/ta1w2rAACgACHH+1rkrbIPAAAA==",
          "result": {
            "__typename": "ApiMedia",
            "id": "QXBpTWVkaWE6DAABCgABHH+1rXDasAAKAAIcf7WuStsg8AAA",
            "media_key": "3_2053559711631585280"
          }
        },
        "media_url_https": "https://pbs.twimg.com/media/HH-1rXDasAAHnCp.jpg",
        "original_info": {
          "focus_rects": [
            {
              "h": 821,
              "w": 1466,
              "x": 0,
              "y": 0
            },
            {
              "h": 1466,
              "w": 1466,
              "x": 0,
              "y": 0
            },
            {
              "h": 1671,
              "w": 1466,
              "x": 0,
              "y": 0
            },
            {
              "h": 1904,
              "w": 952,
              "x": 142,
              "y": 0
            },
            {
              "h": 1904,
              "w": 1466,
              "x": 0,
              "y": 0
            }
          ],
          "height": 1904,
          "width": 1466
        },
        "sizes": {
          "large": {
            "h": 1904,
            "w": 1466
          }
        },
        "type": "photo",
        "url": "https://t.co/JKCt67mfWJ"
      }
    ]
  },
  "card": null,
  "place": {},
  "entities": {
    "hashtags": [],
    "symbols": [],
    "urls": [
      {
        "display_url": "arxiv.org/abs/2604.27233",
        "expanded_url": "https://arxiv.org/abs/2604.27233",
        "indices": [
          1157,
          1180
        ],
        "url": "https://t.co/L0p0UBFcI0"
      },
      {
        "display_url": "academy.dair.ai",
        "expanded_url": "https://academy.dair.ai/",
        "indices": [
          1233,
          1256
        ],
        "url": "https://t.co/1e8RZKs4uX"
      }
    ],
    "user_mentions": []
  },
  "quoted_tweet": null,
  "retweeted_tweet": null,
  "isLimitedReply": false,
  "communityInfo": null,
  "article": null
}