{
  "source_file": "0930",
  "source_platform": "Medium",
  "author": "Benjamin Nweke",
  "title_listed": "Why Most Multi-Agent Systems Fail Even When Evaluation Passes",
  "keywords": "",
  "domain": "大模型算法",
  "content_type": "深度技术解析",
  "is_paid_flag": "否",
  "url": "https://towardsdatascience.com/why-most-multi-agent-systems-fail-even-when-evaluation-passes/",
  "publish_date": "2026-09-07",
  "corpus_stage": "",
  "notes": "",
  "id": "aa9c6400b486",
  "collected_at": "2026-10-02T07:32:47",
  "title": "Why Most Multi-Agent Systems Fail Even When Evaluation Passes | Towards Data Science",
  "body_markdown": "![Revisit consent button](../images/body-df29beacd7496220cc94.svg)\nWe value your privacy\nWe use cookies to enhance your browsing experience, serve personalised ads or content, and analyse our traffic. By clicking \"Accept All\", you consent to our use of cookies.\nCustomise Reject All Accept All\nCustomise Consent Preferences ![](../images/body-2159bb67f3de8ad870eb.svg)\nWe use cookies to help you navigate efficiently and perform certain functions. You will find detailed information about all cookies under each consent category below.\nThe cookies that are categorised as \"Necessary\" are stored on your browser as they are essential for enabling the basic functionalities of the site. ... Show more\nNecessaryAlways Active\nNecessary cookies are required to enable the basic features of this site, such as providing secure log-in or adjusting your consent preferences. These cookies do not store any personally identifiable data.\n  * Cookie\nBCTempID\n  * Duration\n10 minutes\n  * Description\nNo description available.\n\n\n  * Cookie\n__cf_bm\n  * Duration\n1 hour\n  * Description\nThis cookie, set by Cloudflare, is used to support Cloudflare Bot Management. \n\n\n  * Cookie\nAWSALBCORS\n  * Duration\n7 days\n  * Description\nAmazon Web Services set this cookie for load balancing.\n\n\n  * Cookie\n_cfuvid\n  * Duration\nsession\n  * Description\nCloudflare sets this cookie to track users across sessions to optimize user experience by maintaining session consistency and providing personalized services\n\n\n  * Cookie\nli_gc\n  * Duration\n6 months\n  * Description\nLinkedin set this cookie for storing visitor's consent regarding using cookies for non-essential purposes.\n\n\n  * Cookie\n__hssrc\n  * Duration\nsession\n  * Description\nThis cookie is set by Hubspot whenever it changes the session cookie. The __hssrc cookie set to 1 indicates that the user has restarted the browser, and if the cookie does not exist, it is assumed to be a new session.\n\n\n  * Cookie\n__hssc\n  * Duration\n1 hour\n  * Description\nHubSpot sets this cookie to keep track of sessions and to determine if HubSpot should increment the session number and timestamps in the __hstc cookie.\n\n\n  * Cookie\nwpEmojiSettingsSupports\n  * Duration\nsession\n  * Description\nWordPress sets this cookie when a user interacts with emojis on a WordPress site. It helps determine if the user's browser can display emojis properly.\n\n\n  * Cookie\nBCSessionID\n  * Duration\n1 year 1 month 4 days\n  * Description\nBlueconic sets this cookie as a unique identifier for the BlueConic profile.\n\n\n  * Cookie\n_octo\n  * Duration\n1 year\n  * Description\nNo description available.\n\n\n  * Cookie\nlogged_in\n  * Duration\n1 year\n  * Description\nNo description available.\n\n\n  * Cookie\n__Secure-YEC\n  * Duration\npast\n  * Description\nYouTube sets this cookie to stores the user's video player preferences using embedded YouTube video\n\n\n  * Cookie\n__eoi\n  * Duration\n6 months\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nAWSALBTGCORS\n  * Duration\n7 days\n  * Description\nNo description available.\n\n\n  * Cookie\nlogin-status-p\n  * Duration\npast\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nAWSALBTG\n  * Duration\n7 days\n  * Description\nNo description available.\n\n\n  * Cookie\ncsrf_token\n  * Duration\nsession\n  * Description\nNo description available.\n\n\n  * Cookie\ntoken_v2\n  * Duration\n1 day\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nD\n  * Duration\n1 year\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nPHPSESSID\n  * Duration\nsession\n  * Description\nThis cookie is native to PHP applications. The cookie stores and identifies a user's unique session ID to manage user sessions on the website. The cookie is a session cookie and will be deleted when all the browser windows are closed.\n\n\n  * Cookie\nVISITOR_PRIVACY_METADATA\n  * Duration\n6 months\n  * Description\nYouTube sets this cookie to store the user's cookie consent state for the current domain. \n\n\n  * Cookie\ncookietest\n  * Duration\nsession\n  * Description\nThe cookietest cookie is typically used to determine whether the user's browser accepts cookies, essential for website functionality and user experience.\n\n\n  * Cookie\n__Host-airtable-session\n  * Duration\n1 year\n  * Description\nThis cookie is used to enable us to integrate the services of Airtable.\n\n\n  * Cookie\n__Host-airtable-session.sig\n  * Duration\n1 year\n  * Description\nThis cookie is used to enable us to integrate the services of Airtable.\n\n\n  * Cookie\nm\n  * Duration\n1 year 1 month 4 days\n  * Description\nStripe sets this cookie for fraud prevention purposes. It identifies the device used to access the website, allowing the website to be formatted accordingly.\n\n\n  * Cookie\nBIGipServer*\n  * Duration\nsession\n  * Description\nMarketo sets this cookie to collect information about the user's online activity and build a profile about their interests to provide advertisements relevant to the user. \n\n\n  * Cookie\n__cfruid\n  * Duration\nsession\n  * Description\nCloudflare sets this cookie to identify trusted web traffic.\n\n\n  * Cookie\n_GRECAPTCHA\n  * Duration\n6 months\n  * Description\nGoogle Recaptcha service sets this cookie to identify bots to protect the website against malicious spam attacks.\n\n\n  * Cookie\n__Secure-YNID\n  * Duration\n6 months\n  * Description\nGoogle cookie used to protect user security and prevent fraud, especially during the login process.\n\n\n  * Cookie\ncookieyes-consent\n  * Duration\n1 year\n  * Description\nCookieYes sets this cookie to remember users' consent preferences so that their preferences are respected on subsequent visits to this site. It does not collect or store any personal information about the site visitors.\n\n\nFunctional\nFunctional cookies help perform certain functionalities like sharing the content of the website on social media platforms, collecting feedback, and other third-party features.\n  * Cookie\nlidc\n  * Duration\n1 day\n  * Description\nLinkedIn sets the lidc cookie to facilitate data center selection.\n\n\n  * Cookie\nbrw\n  * Duration\n1 year\n  * Description\nNo description available.\n\n\n  * Cookie\nbrwConsent\n  * Duration\n5 minutes\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nWMF-Uniq\n  * Duration\n1 year\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nloom_anon_comment\n  * Duration\n1 year\n  * Description\nNo description available.\n\n\n  * Cookie\nloom_referral_video\n  * Duration\nsession\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nVISITOR_INFO1_LIVE\n  * Duration\n6 months\n  * Description\nA cookie set by YouTube to measure bandwidth that determines whether the user gets the new or old player interface.\n\n\n  * Cookie\nyt-remote-connected-devices\n  * Duration\nNever Expires\n  * Description\nYouTube sets this cookie to store the user's video preferences using embedded YouTube videos.\n\n\n  * Cookie\nytidb::LAST_RESULT_ENTRY_KEY\n  * Duration\nNever Expires\n  * Description\nThe cookie ytidb::LAST_RESULT_ENTRY_KEY is used by YouTube to store the last search result entry that was clicked by the user. This information is used to improve the user experience by providing more relevant search results in the future.\n\n\n  * Cookie\nyt-remote-device-id\n  * Duration\nNever Expires\n  * Description\nYouTube sets this cookie to store the user's video preferences using embedded YouTube videos.\n\n\n  * Cookie\nyt-remote-session-name\n  * Duration\nsession\n  * Description\nThe yt-remote-session-name cookie is used by YouTube to store the user's video player preferences using embedded YouTube video.\n\n\n  * Cookie\nyt-remote-fast-check-period\n  * Duration\nsession\n  * Description\nThe yt-remote-fast-check-period cookie is used by YouTube to store the user's video player preferences for embedded YouTube videos.\n\n\n  * Cookie\nyt-remote-session-app\n  * Duration\nsession\n  * Description\nThe yt-remote-session-app cookie is used by YouTube to store user preferences and information about the interface of the embedded YouTube video player.\n\n\n  * Cookie\nyt-remote-cast-available\n  * Duration\nsession\n  * Description\nThe yt-remote-cast-available cookie is used to store the user's preferences regarding whether casting is available on their YouTube video player.\n\n\n  * Cookie\nyt-remote-cast-installed\n  * Duration\nsession\n  * Description\nThe yt-remote-cast-installed cookie is used to store the user's video player preferences using embedded YouTube video.\n\n\n  * Cookie\ncp_session\n  * Duration\n3 months\n  * Description\nCodepen sets this cookie for Help systems found in the website.\n\n\n  * Cookie\nloid\n  * Duration\n1 year 1 month 4 days\n  * Description\nThis cookie is set by the Reddit. The cookie enables the sharing of content from the website onto the social media platform.\n\n\nAnalytics\nAnalytical cookies are used to understand how visitors interact with the website. These cookies help provide information on metrics such as the number of visitors, bounce rate, traffic source, etc.\n  * Cookie\n__hstc\n  * Duration\n6 months\n  * Description\nHubspot set this main cookie for tracking visitors. It contains the domain, initial timestamp (first visit), last timestamp (last visit), current timestamp (this visit), and session number (increments for each subsequent session).\n\n\n  * Cookie\nhubspotutk\n  * Duration\n6 months\n  * Description\nHubSpot sets this cookie to keep track of the visitors to the website. This cookie is passed to HubSpot on form submission and used when deduplicating contacts.\n\n\n  * Cookie\n_ga\n  * Duration\n1 year 1 month 4 days\n  * Description\nGoogle Analytics sets this cookie to calculate visitor, session and campaign data and track site usage for the site's analytics report. The cookie stores information anonymously and assigns a randomly generated number to recognise unique visitors.\n\n\n  * Cookie\n_ga_*\n  * Duration\n1 year 1 month 4 days\n  * Description\nGoogle Analytics sets this cookie to store and count page views.\n\n\n  * Cookie\n__Host-psifi.analyticsTrace\n  * Duration\n6 hours\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\n__Host-psifi.analyticsTraceV2\n  * Duration\n6 hours\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\n_gh_sess\n  * Duration\nsession\n  * Description\nGitHub sets this cookie for temporary application and framework state between pages like what step the user is on in a multiple step form.\n\n\n  * Cookie\nYSC\n  * Duration\nsession\n  * Description\nYSC cookie is set by Youtube and is used to track the views of embedded videos on Youtube pages.\n\n\n  * Cookie\najs_anonymous_id\n  * Duration\n1 year\n  * Description\nThis cookie is set by Segment to count the number of people who visit a certain site by tracking if they have visited before.\n\n\n  * Cookie\nvuid\n  * Duration\n1 year 1 month 4 days\n  * Description\nVimeo installs this cookie to collect tracking information by setting a unique ID to embed videos on the website. \n\n\nPerformance\nPerformance cookies are used to understand and analyse the key performance indexes of the website which helps in delivering a better user experience for the visitors.\n  * Cookie\nAWSALB\n  * Duration\n7 days\n  * Description\nAWSALB is an application load balancer cookie set by Amazon Web Services to map the session to the target.\n\n\n  * Cookie\nacq\n  * Duration\npast\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nacq.sig\n  * Duration\npast\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nptc\n  * Duration\n2 years\n  * Description\nNo description available.\n\n\nAdvertisement\nAdvertisement cookies are used to provide visitors with customised advertisements based on the pages you visited previously and to analyse the effectiveness of the ad campaigns.\n  * Cookie\nmuc_ads\n  * Duration\n1 year 1 month 4 days\n  * Description\nTwitter sets this cookie to collect user behaviour and interaction data to optimize the website.\n\n\n  * Cookie\nguest_id_marketing\n  * Duration\n1 year 1 month 4 days\n  * Description\nTwitter sets this cookie to identify and track the website visitor.\n\n\n  * Cookie\nguest_id_ads\n  * Duration\n1 year 1 month 4 days\n  * Description\nTwitter sets this cookie to identify and track the website visitor.\n\n\n  * Cookie\npersonalization_id\n  * Duration\n1 year 1 month 4 days\n  * Description\nTwitter sets this cookie to integrate and share features for social media and also store information about how the user uses the website, for tracking and targeting.\n\n\n  * Cookie\nguest_id\n  * Duration\n1 year 1 month 4 days\n  * Description\nTwitter sets this cookie to identify and track the website visitor. It registers if a user is signed in to the Twitter platform and collects information about ad preferences.\n\n\n  * Cookie\nbcookie\n  * Duration\n1 year\n  * Description\nLinkedIn sets this cookie from LinkedIn share buttons and ad tags to recognize browser IDs.\n\n\n  * Cookie\n__Secure-ROLLOUT_TOKEN\n  * Duration\n6 months\n  * Description\nYouTube sets this cookie to manage feature rollout and experimentation. It helps Google control which new features or interface changes are shown to users as part of testing and staged rollouts, ensuring consistent experience for a given user during an experiment.\n\n\n  * Cookie\nyt.innertube::nextId\n  * Duration\nNever Expires\n  * Description\nYouTube sets this cookie to register a unique ID to store data on what videos from YouTube the user has seen.\n\n\n  * Cookie\nyt.innertube::requests\n  * Duration\nNever Expires\n  * Description\nYouTube sets this cookie to register a unique ID to store data on what videos from YouTube the user has seen.\n\n\n  * Cookie\nsession_tracker\n  * Duration\nsession\n  * Description\nThis cookie is set by the Reddit. This cookie is used to identify trusted web traffic. It also helps in adverstising on the website.\n\n\n  * Cookie\nedgebucket\n  * Duration\nsession\n  * Description\nReddit sets this cookie to save the information about a log-on Reddit user, for the purpose of advertisement recommendations and updating the content.\n\n\n  * Cookie\ndid\n  * Duration\n1 year\n  * Description\nArbor sets this cookie to show targeted ads to site visitors.This cookie expires after 2 months or 1 year.\n\n\nUncategorised\nOther uncategorised cookies are those that are being analysed and have not been classified into a category as yet.\nNo cookies to display.\nReject All Save My Preferences Accept All\nToggle Search[](https://towardsdatascience.com/)Toggle Mobile Navigation\n[](https://towardsdatascience.com/)\nPublish AI, ML & data-science insights to a global community of data professionals.\nSign in[Submit an Article](https://write.insightmediagroup.io)\n[Latest](https://towardsdatascience.com/latest/)[Editor's Picks](https://towardsdatascience.com/tag/editors-pick/)[Newsletter](https://towardsdatascience.com/tag/the-variable/)[ShipAI](https://towardsdatascience.com/shipai/)\n* * *\n[Write For TDS](https://towardsdatascience.com/submissions/)\n[LinkedIn](https://www.linkedin.com/company/towards-data-science/)[X](https://x.com/TDataScience)Toggle Search\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\n# Why Most Multi-Agent Systems Fail Even When Evaluation Passes\nHow to catch a payload that looks correct but isn't, using a watchdog pattern with working Python.\n[Benjamin Nweke](https://towardsdatascience.com/author/benjaminnweke/)\nSeptember 7, 202610 min read\n![Two robotic arms pass a glowing golden cube. A crack in the cube reveals chaotic, broken code inside, representing a hidden failure in an otherwise successful agent handoff.](../images/000.webp)\nImage by author (Generated with ChatGPT)\nTwo robotic agents exchange a glowing golden data cube that appears perfect on the outside but contains broken code, illustrating how silent failures can pass unnoticed between agents in a multi-agent system.\nI keep running into some version of this failure wherever agents get chained together, in one shape or another. Take a support-ticket triage system, for example, that is three nodes deep. \nOne classifies the incoming ticket, one pulls the customer's account history from an internal API, and another one drafts the resolution or escalation based on both.\nThen, it ships. It works well in the demo, which honestly, from my experience, doesn't tell you much. It works for the first few days in production too; again, that tells you slightly more but still not enough.\nAlong the line, a complaint comes in about a canceled subscription refund. The account-history node calls the billing API, gets back a 200, and passes the payload downstream as if nothing happened.\nThe payload is empty. No malformed data, no timeout, nothing that would even show up as a 500. Just an empty result set, formatted exactly as a valid response would be, because the account ID got messed up two steps upstream and the billing service quietly returned nothing for an account it couldn't match. \nThe drafting node never sees an error. It sees a well-formed JSON object with no records in it, decides that means \"no billing history,\" and writes a perfectly polite email explaining there's nothing to refund. \nIt goes out, wrong refund decision and all. But still nobody catches it, because nothing about the process ever crashed. As far as the system's concerned, it did its job.\nI don't think this exact scenario needs to have happened to you specifically for it to be worth your time. \nIf you've spent any real stretch of time around multi-agent systems in production, you've either already seen a version of this, or you're going to eventually.\nNone of this is anecdotal, either. [Datadog's 2026 State of AI Engineering report](https://www.datadoghq.com/about/latest-news/press-releases/datadog-state-of-ai-engineering-report-2026/) puts production failure rates for AI requests at around 5 percent, and only about 60 percent of that comes from the loud, capacity-driven failures you'd actually notice through an error code. \nThe rest is closer to what happened above, a request that completes and still gets it wrong.\n![](../images/001.png)\nOne pipeline, three steps, one blind spot. Image by author.\n## Why nothing catches it\nRun this pipeline through a standard evaluation suite, and it sails through. The final output reads well, is grammatically clean, and is professionally worded. \nA human skimming it for tone has no reason to flag anything, not unless they happen to go cross-reference the actual account, which kind of defeats the point of automating the check in the first place. \nIf you score it against a rubric for resolution quality, it probably does well too. Clear, polite, and on topic.\nIt passes because each of those checks looks at the same layer: the final text. None of them ask what happened between node two and node three. \nThe account-history node didn't fail loudly; it failed by succeeding at returning the wrong thing, and succeeding is exactly what output-level eval is built to reward.\nThis is the part I think is worth sitting with longer than feels natural, before jumping to a fix. \nGrading only the compiled output makes you structurally blind to intermediate states that look correct but aren't. That's not a gap you patch with a better prompt on the last node. It's a blind spot that comes baked into where you decided to look in the first place.\n## Grading the UI instead of the application underneath it\nThere's an old comparison here that I keep coming back to. Nobody ships a compiled application and calls it tested because the login screen renders. \nYou test the layer beneath it, the query that backs the login, the token it issues, and even the permission check it triggers along the way. \nThe UI is the last place a bug shows itself, not the first place you'd think to look for one.\nMost production agent eval right now is UI-only testing bolted onto a system that doesn't even have a UI in the traditional sense. \nThe final text response is the only thing getting graded, mostly because it's the only thing that's easy to grade. \nYou can run it through a rubric, compare it line-by-line against a known-good answer, or have someone skim it over coffee.\nThe tool calls, the JSON handoffs between nodes, the partial reasoning getting passed forward: none of that gets watched unless something crashes hard enough to leave a trace in a log somewhere.\nAnd the expensive failure mode was never the loud one. A 500, a server admitting outright that something broke, gets caught and escalated, because the system already expects that shape of failure and has some plan for it. \nThe one that costs you is the 200, the response that says everything's fine, attached to a payload that's structurally fine and semantically garbage.\n## An architecture for watching the middle\nSo the fix isn't another rubric bolted onto the end. It's moving some of the evaluation into the pipeline itself, right at the seams where one agent's output becomes another agent's input.\nI've been calling this an _**Intermediate State Eval architecture**_ , mostly because it needed a name and that one stuck. The idea is a lightweight grader sitting between agent nodes, not waiting for the whole chain to finish. \nIn the ticket-triage case we talked about earlier, that's a checkpoint between the account-history node and the drafting node, and its only job is asking whether the handoff looks plausible. \n> _Does the account ID in the payload actually match the one that was requested?_\n> _Does this look like a real lookup result, or like the kind of default value a system quietly falls back to when it can't find what it's looking for?_\n![](../images/body-68a09cf383089aa9a320.png)\nA watchdog grades each handoff before it's allowed to reach the next node, halting on anything implausible instead of letting it pass through. Image by author.\nYou don't need a large model doing this judgment call, and honestly I'd argue against it. \nA small local model serving as a watchdog between nodes, the same basic idea behind using a [model to judge another model's output](https://arxiv.org/abs/2306.05685), is enough to catch shape-level and plausibility-level problems, and keeping it local prevents the added latency and cost from becoming a second version of the exact problem you're trying to solve.\nIts verdict doesn't actually need to be clever. It just needs to be fast and binary: does this handoff look sane enough to move forward, or does it need to get flagged and stopped before the next node builds anything on top of it? \nThat's the whole job, nothing more.\nHere's what that actually looks like in code. \nStart with the shape of the handoff itself. Defining it as a real schema with [Pydantic](https://docs.pydantic.dev/latest/concepts/models/) instead of a loose dict is what makes the watchdog's job possible at all, since it gives the grader something concrete to check against instead of guessing at structure fresh on every call.\npython\n\n```\nfrom pydantic import BaseModel, Field\nimport logging\n\nlogger = logging.getLogger(\"pipeline.handoff\")\n\n# below this, the watchdog treats a \"plausible\" verdict as implausible anyway -\n# started at 0.5, got burned by a bunch of low-confidence-but-correct handoffs\n# in staging, moved it down until the false-halt rate stopped annoying people\nCONFIDENCE_FLOOR = 0.35\n\n\nclass AccountHistoryPayload(BaseModel):\n    account_id: str\n    subscription_status: str\n    billing_records: list[dict] = Field(default_factory=list)\n    lookup_source: str  # which upstream system actually answered - saved us more than once when this was the only clue\n\n\nclass HandoffVerdict(BaseModel):\n    is_plausible: bool\n    reason: str\n    confidence: float\n```\n\nThe watchdog itself is just a small function, not some new service you need to stand up and maintain. \nIt takes the outgoing payload, the request that produced it, and asks a local model one narrow question instead of an open-ended one. \nKeeping the question narrow is the whole trick, honestly, that's what keeps this cheap enough to actually sit on the critical path instead of becoming its own bottleneck.\npython\n\n```\nimport textwrap\n\n# local_grader: anything callable that takes a prompt and returns text.\n# we're running a distilled 1B model on the same box as the pipeline -\n# it doesn't need to be smart, it needs to be fast and not fall over\n\nGRADER_PROMPT = textwrap.dedent(\"\"\"\\\n    A downstream agent is about to receive this account history payload:\n    {payload}\n\n    It was requested for account_id: {account_id}\n\n    Answer strictly as JSON: {{\"is_plausible\": bool, \"reason\": str, \"confidence\": float}}\n    Flag it as implausible if the account_id doesn't match, if billing_records\n    is empty for a subscription marked active, or if the payload looks like\n    a default/fallback value rather than a real lookup result.\n\"\"\")\n\n\ndef grade_handoff(request_account_id: str, payload: AccountHistoryPayload, local_grader) -> HandoffVerdict:\n    prompt = GRADER_PROMPT.format(payload=payload.model_dump_json(), account_id=request_account_id)\n    raw_response = local_grader(prompt)\n\n    try:\n        verdict = HandoffVerdict.model_validate_json(raw_response)\n    except ValueError:\n        # the grader itself can return garbage - if we can't parse its verdict,\n        # don't just shrug and let the handoff through, that defeats the point\n        logger.error(\"Grader returned unparseable output, blocking handoff: %r\", raw_response[:200])\n        return HandoffVerdict(is_plausible=False, reason=\"grader output unparseable\", confidence=0.0)\n\n    if not verdict.is_plausible or verdict.confidence < CONFIDENCE_FLOOR:\n        logger.warning(\n            \"Handoff rejected for account_id=%s: %s (confidence=%.2f)\\npayload=%s\",\n            request_account_id, verdict.reason, verdict.confidence, payload.model_dump_json(),\n        )\n    return verdict\n```\n\nAnd yeah, the orchestration wiring is about as unglamorous as it gets; that's done on purpose, by the way. Somewhere in your pipeline you already have functions handling classification and drafting the final resolution. \nThe only new piece here is the gate sitting between them, grading the handoff and raising instead of quietly letting a bad payload flow through to whatever writes the customer-facing response.\npython\n\n```\nclass HandoffRejectedError(Exception):\n    pass\n\n\ndef run_pipeline(account_id: str, account_data: dict, local_grader) -> AccountHistoryPayload:\n    payload = AccountHistoryPayload(**account_data)\n    verdict = grade_handoff(account_id, payload, local_grader)\n\n    if not verdict.is_plausible:\n        # this replaces what used to be a wrong email going out quietly -\n        # a raised exception here is annoying, a refund that shouldn't exist isn't\n        raise HandoffRejectedError(f\"account_id={account_id}: {verdict.reason}\")\n\n    return payload  # safe to hand off to whatever drafts the resolution\n```\n\nNone of this is fancy, and I'd be a little suspicious of anyone who dressed it up to sound like it was. \nIt's a schema, one narrow grading call, and an exception where there used to be a silent pass-through. \nThe value was never in the sophistication of any single piece. It's entirely in where you decided to put the check.\nThe practical payoff is that failures stop happening in silence. Instead of a wrong email going out three steps downstream of the real problem, the pipeline stops right at the point of corruption, with the actual bad handoff attached to whatever alert fires. \nThat turns a support escalation that costs you trust into a debugging session that costs you a few minutes.\n## What this costs you\nNone of this is free, and I'd rather be upfront about that than sell you a strict improvement with no downside, because there isn't one, but three costs, specifically:\n  * **Latency:** Plain and simple. For a three-node pipeline, that's two extra inference calls sitting right on the critical path, and it adds up fast if you're building something where response time actually matters.\n  * **A new failure surface:** A miscalibrated watchdog starts rejecting perfectly good handoffs, trading silent corruption for a different annoyance, false halts that somebody now has to triage by hand. Getting that threshold right takes real iteration, not a set-it-once number.\n  * **Judgment:** Deciding where a handoff is actually worth grading, versus where you'd just be adding overhead for its own sake, depends entirely on how expensive being wrong is at that specific point.\n\n\nNot every node needs a watchdog sitting on it. The ones that sit right before something external-facing, an email going out or a refund actually firing, are the ones where the cost of catching a bad handoff clearly beats the cost of the extra hop. \nEverywhere else, you're probably just adding latency for the sake of feeling thorough.\n···\n## Final thoughts\nIf you're running a multi-agent pipeline today and want to try this without tearing the whole thing apart, don't start at the front of the chain. \nTrust me.\nStart at the last internal handoff before something external actually happens, right before an email sends, a record gets written, a decision gets acted on. That's the one boundary worth instrumenting first. \nGive it a week or two, see what the watchdog actually catches, and let that tell you whether it's worth pushing further back into the pipeline.\nYou don't need the full architecture on day one to get something out of this. \nNext time something breaks in a pipeline you own, ask yourself whether your evals would have caught it before the output itself looked wrong. \nMost of the time, if you're honest, the answer's no. That gap is exactly where the riskiest handoff in your system has been sitting the whole time, waiting for someone to look.\nYour final output can lie to you. The trajectory can't.\n···\n## Before you go!\nI write about the engineering decisions that decide whether an AI system holds up in production. You can [subscribe to my newsletter](https://thatcsguy.substack.com/?r=79lldp&utm_campaign=pub-share-checklist) if you'd like more of that.\n**Connect With Me**\n  * [LinkedIn](http://www.linkedin.com/in/nwekebenjamin)\n  * [Substack](https://thatcsguy.substack.com/)\n\n\n## Register Now to Continue Reading\n### Unlock Towards Data Science with a Free Account\nAll articles on Towards Data Science are available to registered users. Create a free account today to access this article and thousands more.\nRegister for freeSign in to your account\n* * *\n![](../images/body-4ad89a2fc5e69dde007d.jpeg)\nWritten By\n## [Benjamin Nweke](https://towardsdatascience.com/author/benjaminnweke/)\nBenjamin Nweke is a Computer Science graduate and technical writer exploring AI, machine learning, and data science. He writes about building reliable AI systems, model evaluation, agentic architectures, and the practical challenges that appear when machine learning moves beyond experiments. His work focuses on making technically complex ideas clear, useful, and grounded in real implementation.\n[See all from Benjamin Nweke](https://towardsdatascience.com/author/benjaminnweke/)\n[AI Agents](https://towardsdatascience.com/tag/ai-agents-2/)[Artificial Intelligence](https://towardsdatascience.com/tag/artificial-intelligence/)[Llm](https://towardsdatascience.com/tag/llm/)[Machine Learning](https://towardsdatascience.com/tag/machine-learning/)[Python](https://towardsdatascience.com/tag/python/)\nShare This Article\n  * [](https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Ftowardsdatascience.com%2Fwhy-most-multi-agent-systems-fail-even-when-evaluation-passes%2F&title=Why%20Most%20Multi-Agent%20Systems%20Fail%20Even%20When%20Evaluation%20Passes)\n  * [](https://www.linkedin.com/shareArticle?mini=true&url=https%3A%2F%2Ftowardsdatascience.com%2Fwhy-most-multi-agent-systems-fail-even-when-evaluation-passes%2F&title=Why%20Most%20Multi-Agent%20Systems%20Fail%20Even%20When%20Evaluation%20Passes)\n  * [](https://x.com/share?url=https%3A%2F%2Ftowardsdatascience.com%2Fwhy-most-multi-agent-systems-fail-even-when-evaluation-passes%2F&text=Why%20Most%20Multi-Agent%20Systems%20Fail%20Even%20When%20Evaluation%20Passes)\n\n\nTowards Data Science is a community publication. Submit your insights to reach our global audience and earn through the TDS Author Payment Program.\n[Write for TDS](https://towardsdatascience.com/submissions/)\n## Related Articles\n  * [![](../images/002.jpg)](https://towardsdatascience.com/how-to-scale-your-llm-usage/)\n## [How to Scale Your LLM Usage](https://towardsdatascience.com/how-to-scale-your-llm-usage/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nLearn how to increase LLM usage to achieve increased productivity\n[Eivind Kjosbakken](https://towardsdatascience.com/author/oieivind/)\nNovember 29, 20256 min read\n  * [![](../images/003.jpg)](https://towardsdatascience.com/agentic-ai-from-first-principles-reflection/)\n## [Agentic AI from First Principles: Reflection](https://towardsdatascience.com/agentic-ai-from-first-principles-reflection/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nFrom theory to code: building feedback loops that improve LLM accuracy\n[Mariya Mansurova](https://towardsdatascience.com/author/miptgirl/)\nOctober 24, 202520 min read\n  * [![](../images/004.jpg)](https://towardsdatascience.com/building-video-game-recommender-systems-with-fastapi-postgresql-and-render-part-1/)\n## [Building Video Game Recommender Systems with FastAPI, PostgreSQL, and Render: Part 1](https://towardsdatascience.com/building-video-game-recommender-systems-with-fastapi-postgresql-and-render-part-1/)\n[Data Science](https://towardsdatascience.com/category/data-science/)\nDesigning a video game recommendations service with Steams API\n[Lucas See](https://towardsdatascience.com/author/seelcs12/)\nSeptember 25, 202521 min read\n  * [![](../images/005.jpg)](https://towardsdatascience.com/automatic-notification-to-email-with-python-810fd357d89c/)\n## [Automatic Notification to Email with Python](https://towardsdatascience.com/automatic-notification-to-email-with-python-810fd357d89c/)\n[Machine Learning](https://towardsdatascience.com/category/artificial-intelligence/machine-learning/)\nAutomate your notification system with Python\n[Cornellius Wijaya](https://towardsdatascience.com/author/cornelliusyudhawijaya/)\nJune 13, 20204 min read\n  * [![](../images/006.jpg)](https://towardsdatascience.com/i-built-an-ai-data-agent-which-can-query-data-and-answer-business-questions-heres-how/)\n## [I Built an AI Data Agent Which Can Query Data and Answer Business Questions. Here’s How.](https://towardsdatascience.com/i-built-an-ai-data-agent-which-can-query-data-and-answer-business-questions-heres-how/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nA step-by-step guide to building a data agent and conversational interface that let business users to explore data in natural language without SQL\n[Jiayan Yin](https://towardsdatascience.com/author/jiayanyin.simba/)\nAugust 6, 202610 min read\n  * [![](../images/007.jpg)](https://towardsdatascience.com/how-to-automate-workflows-with-ai/)\n## [How to Automate Workflows with AI](https://towardsdatascience.com/how-to-automate-workflows-with-ai/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nLearn how to take a manual process and optimize it using AI\n[Eivind Kjosbakken](https://towardsdatascience.com/author/oieivind/)\nNovember 15, 20256 min read\n  * [![](../images/008.jpeg)](https://towardsdatascience.com/bi-is-dead-long-live-bi/)\n## [BI Is Dead, Long Live BI](https://towardsdatascience.com/bi-is-dead-long-live-bi/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nThe true bottleneck was never the analysis.\n[Mahdi Karabiben](https://towardsdatascience.com/author/mahdiqb/)\nJune 11, 20268 min read\n  * [![](../images/009.png)](https://towardsdatascience.com/agentops-is-not-mlops-what-breaks-in-your-monitoring-stack-when-agents-go-to-production/)\n## [AgentOps Is Not MLOps: What Breaks in Your Monitoring Stack When Agents Go to Production](https://towardsdatascience.com/agentops-is-not-mlops-what-breaks-in-your-monitoring-stack-when-agents-go-to-production/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nThe five MLOps monitoring assumptions agents break, and which inherited signals now pass failed runs as healthy.\n[Mostafa Ibrahim](https://towardsdatascience.com/author/mostafaibrahim18/)\nAugust 31, 202610 min read\n  * [![](../images/010.jpg)](https://towardsdatascience.com/deploy-an-openai-agent-builder-chatbot-to-a-website/)\n## [Deploy an OpenAI Agent Builder Chatbot to your Website](https://towardsdatascience.com/deploy-an-openai-agent-builder-chatbot-to-a-website/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nUsing OpenAI’s Agent Builder ChatKit\n[Thomas Reid](https://towardsdatascience.com/author/thomas_reid/)\nOctober 24, 202510 min read\n\n\n  * [YouTube](https://www.youtube.com/c/TowardsDataScience)\n  * [X](https://x.com/TDataScience)\n  * [LinkedIn](https://www.linkedin.com/company/towards-data-science/)\n  * [Threads](https://www.threads.net/@towardsdatascience)\n  * [Bluesky](https://bsky.app/profile/towardsdatascience.com)\n\n\n[](https://towardsdatascience.com/)\nYour home for data science and AI. The world's leading publication for data science, data analytics, data engineering, machine learning, and artificial intelligence professionals.\nSubscribe to Our Newsletter[Write For TDS](https://towardsdatascience.com/submissions/)•[About](https://towardsdatascience.com/about-towards-data-science-d691af11cc2f/)•[Advertise](https://contact.towardsdatascience.com/advertise-with-towards-data-science)•[Privacy Policy](https://towardsdatascience.com/privacy-policy/)•[Terms of Use](https://towardsdatascience.com/website-terms-of-use/)\n© Insight Media Group, LLC 2026•[The New Stack](https://thenewstack.io)•[roadmap.sh](https://roadmap.sh)•[Creator Network](https://insightmediagroup.io/creator-network)\n####  Recommended Articles \nClose\n  * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607686/image?etag=1782161464448&width=500&height=500) How Relevance Models Foreshadowed Transformers for NLP ](https://towardsdatascience.com/how-relevance-models-foreshadowed-transformers-for-nlp/ \"How Relevance Models Foreshadowed Transformers for NLP\")\n  * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607639/image?etag=1781250125848&width=500&height=500) Javascript Fatigue: HTMX Is All You Need to Build ChatGPT — Part 1 ](https://towardsdatascience.com/javascript-fatigue-you-dont-need-js-to-build-chatgpt/ \"Javascript Fatigue: HTMX Is All You Need to Build ChatGPT — Part 1\")\n  * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607930/image?etag=1782124566054&width=500&height=500) The Subset Sum Problem Solved in Linear Time for Dense Enough Inputs ](https://towardsdatascience.com/subset-sum-problem-solved-in-linear-time-for-dense-enough-inputs/ \"The Subset Sum Problem Solved in Linear Time for Dense Enough Inputs\")\n  * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/3660/image?etag=1780176246070&width=500&height=500) Does Your Company Have a Data Strategy? ](https://towardsdatascience.com/does-your-company-have-a-data-strategy-e8944dbec4ae/ \"Does Your Company Have a Data Strategy?\")\n\n\n",
  "metadata": {
    "title": "Why Most Multi-Agent Systems Fail Even When Evaluation Passes | Towards Data Science",
    "description": "How to catch a payload that looks correct but isn't, using a watchdog pattern with working Python.",
    "keywords": null,
    "author": "Benjamin Nweke",
    "og:locale": "en_US",
    "og:type": "article",
    "og:title": "Why Most Multi-Agent Systems Fail Even When Evaluation Passes",
    "og:description": "How to catch a payload that looks correct but isn't, using a watchdog pattern with working Python.",
    "og:url": "https://towardsdatascience.com/why-most-multi-agent-systems-fail-even-when-evaluation-passes/",
    "og:site_name": "Towards Data Science",
    "og:image": "https://assets.insightmediagroup.io/media/1788472869074_98pjgr.webp",
    "og:image:width": "1402",
    "og:image:height": "1122",
    "og:image:type": "image/webp",
    "twitter:card": "summary_large_image",
    "twitter:site": "@TDataScience",
    "twitter:creator": "@TDataScience",
    "twitter:label1": "Written by",
    "twitter:data1": "Benjamin Nweke",
    "twitter:label2": "Est. reading time",
    "twitter:data2": "10 min read",
    "article:published_time": "2026-09-07T12:00:01.956Z",
    "article:modified_time": "2026-09-07T12:00:01.956Z",
    "article:author": "https://towardsdatascience.com/author/benjaminnweke/"
  },
  "paywall": {
    "listed_paid": false,
    "detected": false,
    "marker": "",
    "public_only": false,
    "policy": "付费/会员内容只保留页面公开可见部分，不绕过付费墙"
  },
  "crawl": {
    "success": true,
    "status_code": 200,
    "error": ""
  },
  "code_blocks": [
    {
      "language": "txt",
      "file": "attachments/aa9c6400b486/code/block_000.txt",
      "chars": 732
    },
    {
      "language": "txt",
      "file": "attachments/aa9c6400b486/code/block_001.txt",
      "chars": 1738
    },
    {
      "language": "txt",
      "file": "attachments/aa9c6400b486/code/block_002.txt",
      "chars": 605
    }
  ],
  "images": [
    {
      "status": "http_404",
      "url": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607686/image?etag=1782161464448&width=500&height=500",
      "src": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607686/image?etag=1782161464448&width=500&height=500",
      "alt": "",
      "downloaded": false,
      "discovery": "saved_markdown",
      "retried_at": "2026-10-20T18:04:54Z"
    },
    {
      "status": "http_404",
      "url": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607639/image?etag=1781250125848&width=500&height=500",
      "src": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607639/image?etag=1781250125848&width=500&height=500",
      "alt": "",
      "downloaded": false,
      "discovery": "saved_markdown",
      "retried_at": "2026-10-20T18:04:54Z"
    },
    {
      "status": "http_404",
      "url": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607930/image?etag=1782124566054&width=500&height=500",
      "src": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607930/image?etag=1782124566054&width=500&height=500",
      "alt": "",
      "downloaded": false,
      "discovery": "saved_markdown",
      "retried_at": "2026-10-20T18:04:54Z"
    },
    {
      "status": "http_404",
      "url": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/3660/image?etag=1780176246070&width=500&height=500",
      "src": "https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/3660/image?etag=1780176246070&width=500&height=500",
      "alt": "",
      "downloaded": false,
      "discovery": "saved_markdown",
      "retried_at": "2026-10-20T18:04:54Z"
    }
  ],
  "audio": [],
  "videos": [
    {
      "url": "https://www.youtube.com/c/TowardsDataScience",
      "kind": "channel_reference",
      "status": "reference_only"
    }
  ],
  "links": [
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/latest/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/tag/editors-pick/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/tag/the-variable/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/shipai/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/submissions/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/author/benjaminnweke/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/tag/ai-agents-2/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/tag/artificial-intelligence/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/tag/llm/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/tag/machine-learning/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/tag/python/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/how-to-scale-your-llm-usage/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/author/oieivind/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/agentic-ai-from-first-principles-reflection/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/author/miptgirl/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/building-video-game-recommender-systems-with-fastapi-postgresql-and-render-part-1/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/category/data-science/"
    },
    {
      "type": "internal",
      "href": "https://towardsdatascience.com/author/seelcs12/"
    },
    {
      "type": "external",
      "href": "https://write.insightmediagroup.io"
    },
    {
      "type": "external",
      "href": "https://www.linkedin.com/company/towards-data-science/"
    },
    {
      "type": "external",
      "href": "https://x.com/TDataScience"
    },
    {
      "type": "external",
      "href": "https://www.datadoghq.com/about/latest-news/press-releases/datadog-state-of-ai-engineering-report-2026/"
    },
    {
      "type": "external",
      "href": "https://arxiv.org/abs/2306.05685"
    },
    {
      "type": "external",
      "href": "https://docs.pydantic.dev/latest/concepts/models/"
    },
    {
      "type": "external",
      "href": "https://thatcsguy.substack.com/?r=79lldp"
    },
    {
      "type": "external",
      "href": "http://www.linkedin.com/in/nwekebenjamin"
    },
    {
      "type": "external",
      "href": "https://thatcsguy.substack.com/"
    },
    {
      "type": "external",
      "href": "https://www.facebook.com/sharer/sharer.php?title=Why+Most+Multi-Agent+Systems+Fail+Even+When+Evaluation+Passes&u=https%3A%2F%2Ftowardsdatascience.com%2Fwhy-most-multi-agent-systems-fail-even-when-evaluation-passes%2F"
    },
    {
      "type": "external",
      "href": "https://www.linkedin.com/shareArticle?mini=true&title=Why+Most+Multi-Agent+Systems+Fail+Even+When+Evaluation+Passes&url=https%3A%2F%2Ftowardsdatascience.com%2Fwhy-most-multi-agent-systems-fail-even-when-evaluation-passes%2F"
    },
    {
      "type": "external",
      "href": "https://x.com/share?text=Why+Most+Multi-Agent+Systems+Fail+Even+When+Evaluation+Passes&url=https%3A%2F%2Ftowardsdatascience.com%2Fwhy-most-multi-agent-systems-fail-even-when-evaluation-passes%2F"
    },
    {
      "type": "external",
      "href": "https://www.youtube.com/c/TowardsDataScience"
    },
    {
      "type": "external",
      "href": "https://www.threads.net/@towardsdatascience"
    },
    {
      "type": "external",
      "href": "https://bsky.app/profile/towardsdatascience.com"
    },
    {
      "type": "external",
      "href": "https://thenewstack.io"
    },
    {
      "type": "external",
      "href": "https://roadmap.sh"
    },
    {
      "type": "external",
      "href": "https://insightmediagroup.io/creator-network"
    }
  ],
  "repair_version": 2,
  "repaired_at": "2026-10-20T17:31:52Z",
  "source_url": "https://towardsdatascience.com/why-most-multi-agent-systems-fail-even-when-evaluation-passes/",
  "delivery_status": "unverified",
  "discovery_status": "http_403",
  "missing_reason": "源页面未重新核验:http_403；已按保留正文补采图片",
  "repair_error": "",
  "actual_video_files": [],
  "expected_saved_body_images": 4,
  "image_discovery": "saved_markdown",
  "source_body_markdown": "![Revisit consent button](https://cdn-cookieyes.com/assets/images/revisit.svg)\nWe value your privacy\nWe use cookies to enhance your browsing experience, serve personalised ads or content, and analyse our traffic. By clicking \"Accept All\", you consent to our use of cookies.\nCustomise Reject All Accept All\nCustomise Consent Preferences ![](https://cdn-cookieyes.com/assets/images/close.svg)\nWe use cookies to help you navigate efficiently and perform certain functions. You will find detailed information about all cookies under each consent category below.\nThe cookies that are categorised as \"Necessary\" are stored on your browser as they are essential for enabling the basic functionalities of the site. ... Show more\nNecessaryAlways Active\nNecessary cookies are required to enable the basic features of this site, such as providing secure log-in or adjusting your consent preferences. These cookies do not store any personally identifiable data.\n  * Cookie\nBCTempID\n  * Duration\n10 minutes\n  * Description\nNo description available.\n\n\n  * Cookie\n__cf_bm\n  * Duration\n1 hour\n  * Description\nThis cookie, set by Cloudflare, is used to support Cloudflare Bot Management. \n\n\n  * Cookie\nAWSALBCORS\n  * Duration\n7 days\n  * Description\nAmazon Web Services set this cookie for load balancing.\n\n\n  * Cookie\n_cfuvid\n  * Duration\nsession\n  * Description\nCloudflare sets this cookie to track users across sessions to optimize user experience by maintaining session consistency and providing personalized services\n\n\n  * Cookie\nli_gc\n  * Duration\n6 months\n  * Description\nLinkedin set this cookie for storing visitor's consent regarding using cookies for non-essential purposes.\n\n\n  * Cookie\n__hssrc\n  * Duration\nsession\n  * Description\nThis cookie is set by Hubspot whenever it changes the session cookie. The __hssrc cookie set to 1 indicates that the user has restarted the browser, and if the cookie does not exist, it is assumed to be a new session.\n\n\n  * Cookie\n__hssc\n  * Duration\n1 hour\n  * Description\nHubSpot sets this cookie to keep track of sessions and to determine if HubSpot should increment the session number and timestamps in the __hstc cookie.\n\n\n  * Cookie\nwpEmojiSettingsSupports\n  * Duration\nsession\n  * Description\nWordPress sets this cookie when a user interacts with emojis on a WordPress site. It helps determine if the user's browser can display emojis properly.\n\n\n  * Cookie\nBCSessionID\n  * Duration\n1 year 1 month 4 days\n  * Description\nBlueconic sets this cookie as a unique identifier for the BlueConic profile.\n\n\n  * Cookie\n_octo\n  * Duration\n1 year\n  * Description\nNo description available.\n\n\n  * Cookie\nlogged_in\n  * Duration\n1 year\n  * Description\nNo description available.\n\n\n  * Cookie\n__Secure-YEC\n  * Duration\npast\n  * Description\nYouTube sets this cookie to stores the user's video player preferences using embedded YouTube video\n\n\n  * Cookie\n__eoi\n  * Duration\n6 months\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nAWSALBTGCORS\n  * Duration\n7 days\n  * Description\nNo description available.\n\n\n  * Cookie\nlogin-status-p\n  * Duration\npast\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nAWSALBTG\n  * Duration\n7 days\n  * Description\nNo description available.\n\n\n  * Cookie\ncsrf_token\n  * Duration\nsession\n  * Description\nNo description available.\n\n\n  * Cookie\ntoken_v2\n  * Duration\n1 day\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nD\n  * Duration\n1 year\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nPHPSESSID\n  * Duration\nsession\n  * Description\nThis cookie is native to PHP applications. The cookie stores and identifies a user's unique session ID to manage user sessions on the website. The cookie is a session cookie and will be deleted when all the browser windows are closed.\n\n\n  * Cookie\nVISITOR_PRIVACY_METADATA\n  * Duration\n6 months\n  * Description\nYouTube sets this cookie to store the user's cookie consent state for the current domain. \n\n\n  * Cookie\ncookietest\n  * Duration\nsession\n  * Description\nThe cookietest cookie is typically used to determine whether the user's browser accepts cookies, essential for website functionality and user experience.\n\n\n  * Cookie\n__Host-airtable-session\n  * Duration\n1 year\n  * Description\nThis cookie is used to enable us to integrate the services of Airtable.\n\n\n  * Cookie\n__Host-airtable-session.sig\n  * Duration\n1 year\n  * Description\nThis cookie is used to enable us to integrate the services of Airtable.\n\n\n  * Cookie\nm\n  * Duration\n1 year 1 month 4 days\n  * Description\nStripe sets this cookie for fraud prevention purposes. It identifies the device used to access the website, allowing the website to be formatted accordingly.\n\n\n  * Cookie\nBIGipServer*\n  * Duration\nsession\n  * Description\nMarketo sets this cookie to collect information about the user's online activity and build a profile about their interests to provide advertisements relevant to the user. \n\n\n  * Cookie\n__cfruid\n  * Duration\nsession\n  * Description\nCloudflare sets this cookie to identify trusted web traffic.\n\n\n  * Cookie\n_GRECAPTCHA\n  * Duration\n6 months\n  * Description\nGoogle Recaptcha service sets this cookie to identify bots to protect the website against malicious spam attacks.\n\n\n  * Cookie\n__Secure-YNID\n  * Duration\n6 months\n  * Description\nGoogle cookie used to protect user security and prevent fraud, especially during the login process.\n\n\n  * Cookie\ncookieyes-consent\n  * Duration\n1 year\n  * Description\nCookieYes sets this cookie to remember users' consent preferences so that their preferences are respected on subsequent visits to this site. It does not collect or store any personal information about the site visitors.\n\n\nFunctional\nFunctional cookies help perform certain functionalities like sharing the content of the website on social media platforms, collecting feedback, and other third-party features.\n  * Cookie\nlidc\n  * Duration\n1 day\n  * Description\nLinkedIn sets the lidc cookie to facilitate data center selection.\n\n\n  * Cookie\nbrw\n  * Duration\n1 year\n  * Description\nNo description available.\n\n\n  * Cookie\nbrwConsent\n  * Duration\n5 minutes\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nWMF-Uniq\n  * Duration\n1 year\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nloom_anon_comment\n  * Duration\n1 year\n  * Description\nNo description available.\n\n\n  * Cookie\nloom_referral_video\n  * Duration\nsession\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nVISITOR_INFO1_LIVE\n  * Duration\n6 months\n  * Description\nA cookie set by YouTube to measure bandwidth that determines whether the user gets the new or old player interface.\n\n\n  * Cookie\nyt-remote-connected-devices\n  * Duration\nNever Expires\n  * Description\nYouTube sets this cookie to store the user's video preferences using embedded YouTube videos.\n\n\n  * Cookie\nytidb::LAST_RESULT_ENTRY_KEY\n  * Duration\nNever Expires\n  * Description\nThe cookie ytidb::LAST_RESULT_ENTRY_KEY is used by YouTube to store the last search result entry that was clicked by the user. This information is used to improve the user experience by providing more relevant search results in the future.\n\n\n  * Cookie\nyt-remote-device-id\n  * Duration\nNever Expires\n  * Description\nYouTube sets this cookie to store the user's video preferences using embedded YouTube videos.\n\n\n  * Cookie\nyt-remote-session-name\n  * Duration\nsession\n  * Description\nThe yt-remote-session-name cookie is used by YouTube to store the user's video player preferences using embedded YouTube video.\n\n\n  * Cookie\nyt-remote-fast-check-period\n  * Duration\nsession\n  * Description\nThe yt-remote-fast-check-period cookie is used by YouTube to store the user's video player preferences for embedded YouTube videos.\n\n\n  * Cookie\nyt-remote-session-app\n  * Duration\nsession\n  * Description\nThe yt-remote-session-app cookie is used by YouTube to store user preferences and information about the interface of the embedded YouTube video player.\n\n\n  * Cookie\nyt-remote-cast-available\n  * Duration\nsession\n  * Description\nThe yt-remote-cast-available cookie is used to store the user's preferences regarding whether casting is available on their YouTube video player.\n\n\n  * Cookie\nyt-remote-cast-installed\n  * Duration\nsession\n  * Description\nThe yt-remote-cast-installed cookie is used to store the user's video player preferences using embedded YouTube video.\n\n\n  * Cookie\ncp_session\n  * Duration\n3 months\n  * Description\nCodepen sets this cookie for Help systems found in the website.\n\n\n  * Cookie\nloid\n  * Duration\n1 year 1 month 4 days\n  * Description\nThis cookie is set by the Reddit. The cookie enables the sharing of content from the website onto the social media platform.\n\n\nAnalytics\nAnalytical cookies are used to understand how visitors interact with the website. These cookies help provide information on metrics such as the number of visitors, bounce rate, traffic source, etc.\n  * Cookie\n__hstc\n  * Duration\n6 months\n  * Description\nHubspot set this main cookie for tracking visitors. It contains the domain, initial timestamp (first visit), last timestamp (last visit), current timestamp (this visit), and session number (increments for each subsequent session).\n\n\n  * Cookie\nhubspotutk\n  * Duration\n6 months\n  * Description\nHubSpot sets this cookie to keep track of the visitors to the website. This cookie is passed to HubSpot on form submission and used when deduplicating contacts.\n\n\n  * Cookie\n_ga\n  * Duration\n1 year 1 month 4 days\n  * Description\nGoogle Analytics sets this cookie to calculate visitor, session and campaign data and track site usage for the site's analytics report. The cookie stores information anonymously and assigns a randomly generated number to recognise unique visitors.\n\n\n  * Cookie\n_ga_*\n  * Duration\n1 year 1 month 4 days\n  * Description\nGoogle Analytics sets this cookie to store and count page views.\n\n\n  * Cookie\n__Host-psifi.analyticsTrace\n  * Duration\n6 hours\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\n__Host-psifi.analyticsTraceV2\n  * Duration\n6 hours\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\n_gh_sess\n  * Duration\nsession\n  * Description\nGitHub sets this cookie for temporary application and framework state between pages like what step the user is on in a multiple step form.\n\n\n  * Cookie\nYSC\n  * Duration\nsession\n  * Description\nYSC cookie is set by Youtube and is used to track the views of embedded videos on Youtube pages.\n\n\n  * Cookie\najs_anonymous_id\n  * Duration\n1 year\n  * Description\nThis cookie is set by Segment to count the number of people who visit a certain site by tracking if they have visited before.\n\n\n  * Cookie\nvuid\n  * Duration\n1 year 1 month 4 days\n  * Description\nVimeo installs this cookie to collect tracking information by setting a unique ID to embed videos on the website. \n\n\nPerformance\nPerformance cookies are used to understand and analyse the key performance indexes of the website which helps in delivering a better user experience for the visitors.\n  * Cookie\nAWSALB\n  * Duration\n7 days\n  * Description\nAWSALB is an application load balancer cookie set by Amazon Web Services to map the session to the target.\n\n\n  * Cookie\nacq\n  * Duration\npast\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nacq.sig\n  * Duration\npast\n  * Description\nDescription is currently not available.\n\n\n  * Cookie\nptc\n  * Duration\n2 years\n  * Description\nNo description available.\n\n\nAdvertisement\nAdvertisement cookies are used to provide visitors with customised advertisements based on the pages you visited previously and to analyse the effectiveness of the ad campaigns.\n  * Cookie\nmuc_ads\n  * Duration\n1 year 1 month 4 days\n  * Description\nTwitter sets this cookie to collect user behaviour and interaction data to optimize the website.\n\n\n  * Cookie\nguest_id_marketing\n  * Duration\n1 year 1 month 4 days\n  * Description\nTwitter sets this cookie to identify and track the website visitor.\n\n\n  * Cookie\nguest_id_ads\n  * Duration\n1 year 1 month 4 days\n  * Description\nTwitter sets this cookie to identify and track the website visitor.\n\n\n  * Cookie\npersonalization_id\n  * Duration\n1 year 1 month 4 days\n  * Description\nTwitter sets this cookie to integrate and share features for social media and also store information about how the user uses the website, for tracking and targeting.\n\n\n  * Cookie\nguest_id\n  * Duration\n1 year 1 month 4 days\n  * Description\nTwitter sets this cookie to identify and track the website visitor. It registers if a user is signed in to the Twitter platform and collects information about ad preferences.\n\n\n  * Cookie\nbcookie\n  * Duration\n1 year\n  * Description\nLinkedIn sets this cookie from LinkedIn share buttons and ad tags to recognize browser IDs.\n\n\n  * Cookie\n__Secure-ROLLOUT_TOKEN\n  * Duration\n6 months\n  * Description\nYouTube sets this cookie to manage feature rollout and experimentation. It helps Google control which new features or interface changes are shown to users as part of testing and staged rollouts, ensuring consistent experience for a given user during an experiment.\n\n\n  * Cookie\nyt.innertube::nextId\n  * Duration\nNever Expires\n  * Description\nYouTube sets this cookie to register a unique ID to store data on what videos from YouTube the user has seen.\n\n\n  * Cookie\nyt.innertube::requests\n  * Duration\nNever Expires\n  * Description\nYouTube sets this cookie to register a unique ID to store data on what videos from YouTube the user has seen.\n\n\n  * Cookie\nsession_tracker\n  * Duration\nsession\n  * Description\nThis cookie is set by the Reddit. This cookie is used to identify trusted web traffic. It also helps in adverstising on the website.\n\n\n  * Cookie\nedgebucket\n  * Duration\nsession\n  * Description\nReddit sets this cookie to save the information about a log-on Reddit user, for the purpose of advertisement recommendations and updating the content.\n\n\n  * Cookie\ndid\n  * Duration\n1 year\n  * Description\nArbor sets this cookie to show targeted ads to site visitors.This cookie expires after 2 months or 1 year.\n\n\nUncategorised\nOther uncategorised cookies are those that are being analysed and have not been classified into a category as yet.\nNo cookies to display.\nReject All Save My Preferences Accept All\nToggle Search[](https://towardsdatascience.com/)Toggle Mobile Navigation\n[](https://towardsdatascience.com/)\nPublish AI, ML & data-science insights to a global community of data professionals.\nSign in[Submit an Article](https://write.insightmediagroup.io)\n[Latest](https://towardsdatascience.com/latest/)[Editor's Picks](https://towardsdatascience.com/tag/editors-pick/)[Newsletter](https://towardsdatascience.com/tag/the-variable/)[ShipAI](https://towardsdatascience.com/shipai/)\n* * *\n[Write For TDS](https://towardsdatascience.com/submissions/)\n[LinkedIn](https://www.linkedin.com/company/towards-data-science/)[X](https://x.com/TDataScience)Toggle Search\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\n# Why Most Multi-Agent Systems Fail Even When Evaluation Passes\nHow to catch a payload that looks correct but isn't, using a watchdog pattern with working Python.\n[Benjamin Nweke](https://towardsdatascience.com/author/benjaminnweke/)\nSeptember 7, 202610 min read\n![Two robotic arms pass a glowing golden cube. A crack in the cube reveals chaotic, broken code inside, representing a hidden failure in an otherwise successful agent handoff.](https://assets.insightmediagroup.io/media/1788472869074_98pjgr.webp)\nImage by author (Generated with ChatGPT)\nTwo robotic agents exchange a glowing golden data cube that appears perfect on the outside but contains broken code, illustrating how silent failures can pass unnoticed between agents in a multi-agent system.\nI keep running into some version of this failure wherever agents get chained together, in one shape or another. Take a support-ticket triage system, for example, that is three nodes deep. \nOne classifies the incoming ticket, one pulls the customer's account history from an internal API, and another one drafts the resolution or escalation based on both.\nThen, it ships. It works well in the demo, which honestly, from my experience, doesn't tell you much. It works for the first few days in production too; again, that tells you slightly more but still not enough.\nAlong the line, a complaint comes in about a canceled subscription refund. The account-history node calls the billing API, gets back a 200, and passes the payload downstream as if nothing happened.\nThe payload is empty. No malformed data, no timeout, nothing that would even show up as a 500. Just an empty result set, formatted exactly as a valid response would be, because the account ID got messed up two steps upstream and the billing service quietly returned nothing for an account it couldn't match. \nThe drafting node never sees an error. It sees a well-formed JSON object with no records in it, decides that means \"no billing history,\" and writes a perfectly polite email explaining there's nothing to refund. \nIt goes out, wrong refund decision and all. But still nobody catches it, because nothing about the process ever crashed. As far as the system's concerned, it did its job.\nI don't think this exact scenario needs to have happened to you specifically for it to be worth your time. \nIf you've spent any real stretch of time around multi-agent systems in production, you've either already seen a version of this, or you're going to eventually.\nNone of this is anecdotal, either. [Datadog's 2026 State of AI Engineering report](https://www.datadoghq.com/about/latest-news/press-releases/datadog-state-of-ai-engineering-report-2026/) puts production failure rates for AI requests at around 5 percent, and only about 60 percent of that comes from the loud, capacity-driven failures you'd actually notice through an error code. \nThe rest is closer to what happened above, a request that completes and still gets it wrong.\n![](https://assets.insightmediagroup.io/media/1788421524643_lcdguw.png)\nOne pipeline, three steps, one blind spot. Image by author.\n## Why nothing catches it\nRun this pipeline through a standard evaluation suite, and it sails through. The final output reads well, is grammatically clean, and is professionally worded. \nA human skimming it for tone has no reason to flag anything, not unless they happen to go cross-reference the actual account, which kind of defeats the point of automating the check in the first place. \nIf you score it against a rubric for resolution quality, it probably does well too. Clear, polite, and on topic.\nIt passes because each of those checks looks at the same layer: the final text. None of them ask what happened between node two and node three. \nThe account-history node didn't fail loudly; it failed by succeeding at returning the wrong thing, and succeeding is exactly what output-level eval is built to reward.\nThis is the part I think is worth sitting with longer than feels natural, before jumping to a fix. \nGrading only the compiled output makes you structurally blind to intermediate states that look correct but aren't. That's not a gap you patch with a better prompt on the last node. It's a blind spot that comes baked into where you decided to look in the first place.\n## Grading the UI instead of the application underneath it\nThere's an old comparison here that I keep coming back to. Nobody ships a compiled application and calls it tested because the login screen renders. \nYou test the layer beneath it, the query that backs the login, the token it issues, and even the permission check it triggers along the way. \nThe UI is the last place a bug shows itself, not the first place you'd think to look for one.\nMost production agent eval right now is UI-only testing bolted onto a system that doesn't even have a UI in the traditional sense. \nThe final text response is the only thing getting graded, mostly because it's the only thing that's easy to grade. \nYou can run it through a rubric, compare it line-by-line against a known-good answer, or have someone skim it over coffee.\nThe tool calls, the JSON handoffs between nodes, the partial reasoning getting passed forward: none of that gets watched unless something crashes hard enough to leave a trace in a log somewhere.\nAnd the expensive failure mode was never the loud one. A 500, a server admitting outright that something broke, gets caught and escalated, because the system already expects that shape of failure and has some plan for it. \nThe one that costs you is the 200, the response that says everything's fine, attached to a payload that's structurally fine and semantically garbage.\n## An architecture for watching the middle\nSo the fix isn't another rubric bolted onto the end. It's moving some of the evaluation into the pipeline itself, right at the seams where one agent's output becomes another agent's input.\nI've been calling this an _**Intermediate State Eval architecture**_ , mostly because it needed a name and that one stuck. The idea is a lightweight grader sitting between agent nodes, not waiting for the whole chain to finish. \nIn the ticket-triage case we talked about earlier, that's a checkpoint between the account-history node and the drafting node, and its only job is asking whether the handoff looks plausible. \n> _Does the account ID in the payload actually match the one that was requested?_\n> _Does this look like a real lookup result, or like the kind of default value a system quietly falls back to when it can't find what it's looking for?_\n![](https://assets.insightmediagroup.io/media/1788294399527_dh9qt4.png)\nA watchdog grades each handoff before it's allowed to reach the next node, halting on anything implausible instead of letting it pass through. Image by author.\nYou don't need a large model doing this judgment call, and honestly I'd argue against it. \nA small local model serving as a watchdog between nodes, the same basic idea behind using a [model to judge another model's output](https://arxiv.org/abs/2306.05685), is enough to catch shape-level and plausibility-level problems, and keeping it local prevents the added latency and cost from becoming a second version of the exact problem you're trying to solve.\nIts verdict doesn't actually need to be clever. It just needs to be fast and binary: does this handoff look sane enough to move forward, or does it need to get flagged and stopped before the next node builds anything on top of it? \nThat's the whole job, nothing more.\nHere's what that actually looks like in code. \nStart with the shape of the handoff itself. Defining it as a real schema with [Pydantic](https://docs.pydantic.dev/latest/concepts/models/) instead of a loose dict is what makes the watchdog's job possible at all, since it gives the grader something concrete to check against instead of guessing at structure fresh on every call.\npython\n\n```\nfrom pydantic import BaseModel, Field\nimport logging\n\nlogger = logging.getLogger(\"pipeline.handoff\")\n\n# below this, the watchdog treats a \"plausible\" verdict as implausible anyway -\n# started at 0.5, got burned by a bunch of low-confidence-but-correct handoffs\n# in staging, moved it down until the false-halt rate stopped annoying people\nCONFIDENCE_FLOOR = 0.35\n\n\nclass AccountHistoryPayload(BaseModel):\n    account_id: str\n    subscription_status: str\n    billing_records: list[dict] = Field(default_factory=list)\n    lookup_source: str  # which upstream system actually answered - saved us more than once when this was the only clue\n\n\nclass HandoffVerdict(BaseModel):\n    is_plausible: bool\n    reason: str\n    confidence: float\n```\n\nThe watchdog itself is just a small function, not some new service you need to stand up and maintain. \nIt takes the outgoing payload, the request that produced it, and asks a local model one narrow question instead of an open-ended one. \nKeeping the question narrow is the whole trick, honestly, that's what keeps this cheap enough to actually sit on the critical path instead of becoming its own bottleneck.\npython\n\n```\nimport textwrap\n\n# local_grader: anything callable that takes a prompt and returns text.\n# we're running a distilled 1B model on the same box as the pipeline -\n# it doesn't need to be smart, it needs to be fast and not fall over\n\nGRADER_PROMPT = textwrap.dedent(\"\"\"\\\n    A downstream agent is about to receive this account history payload:\n    {payload}\n\n    It was requested for account_id: {account_id}\n\n    Answer strictly as JSON: {{\"is_plausible\": bool, \"reason\": str, \"confidence\": float}}\n    Flag it as implausible if the account_id doesn't match, if billing_records\n    is empty for a subscription marked active, or if the payload looks like\n    a default/fallback value rather than a real lookup result.\n\"\"\")\n\n\ndef grade_handoff(request_account_id: str, payload: AccountHistoryPayload, local_grader) -> HandoffVerdict:\n    prompt = GRADER_PROMPT.format(payload=payload.model_dump_json(), account_id=request_account_id)\n    raw_response = local_grader(prompt)\n\n    try:\n        verdict = HandoffVerdict.model_validate_json(raw_response)\n    except ValueError:\n        # the grader itself can return garbage - if we can't parse its verdict,\n        # don't just shrug and let the handoff through, that defeats the point\n        logger.error(\"Grader returned unparseable output, blocking handoff: %r\", raw_response[:200])\n        return HandoffVerdict(is_plausible=False, reason=\"grader output unparseable\", confidence=0.0)\n\n    if not verdict.is_plausible or verdict.confidence < CONFIDENCE_FLOOR:\n        logger.warning(\n            \"Handoff rejected for account_id=%s: %s (confidence=%.2f)\\npayload=%s\",\n            request_account_id, verdict.reason, verdict.confidence, payload.model_dump_json(),\n        )\n    return verdict\n```\n\nAnd yeah, the orchestration wiring is about as unglamorous as it gets; that's done on purpose, by the way. Somewhere in your pipeline you already have functions handling classification and drafting the final resolution. \nThe only new piece here is the gate sitting between them, grading the handoff and raising instead of quietly letting a bad payload flow through to whatever writes the customer-facing response.\npython\n\n```\nclass HandoffRejectedError(Exception):\n    pass\n\n\ndef run_pipeline(account_id: str, account_data: dict, local_grader) -> AccountHistoryPayload:\n    payload = AccountHistoryPayload(**account_data)\n    verdict = grade_handoff(account_id, payload, local_grader)\n\n    if not verdict.is_plausible:\n        # this replaces what used to be a wrong email going out quietly -\n        # a raised exception here is annoying, a refund that shouldn't exist isn't\n        raise HandoffRejectedError(f\"account_id={account_id}: {verdict.reason}\")\n\n    return payload  # safe to hand off to whatever drafts the resolution\n```\n\nNone of this is fancy, and I'd be a little suspicious of anyone who dressed it up to sound like it was. \nIt's a schema, one narrow grading call, and an exception where there used to be a silent pass-through. \nThe value was never in the sophistication of any single piece. It's entirely in where you decided to put the check.\nThe practical payoff is that failures stop happening in silence. Instead of a wrong email going out three steps downstream of the real problem, the pipeline stops right at the point of corruption, with the actual bad handoff attached to whatever alert fires. \nThat turns a support escalation that costs you trust into a debugging session that costs you a few minutes.\n## What this costs you\nNone of this is free, and I'd rather be upfront about that than sell you a strict improvement with no downside, because there isn't one, but three costs, specifically:\n  * **Latency:** Plain and simple. For a three-node pipeline, that's two extra inference calls sitting right on the critical path, and it adds up fast if you're building something where response time actually matters.\n  * **A new failure surface:** A miscalibrated watchdog starts rejecting perfectly good handoffs, trading silent corruption for a different annoyance, false halts that somebody now has to triage by hand. Getting that threshold right takes real iteration, not a set-it-once number.\n  * **Judgment:** Deciding where a handoff is actually worth grading, versus where you'd just be adding overhead for its own sake, depends entirely on how expensive being wrong is at that specific point.\n\n\nNot every node needs a watchdog sitting on it. The ones that sit right before something external-facing, an email going out or a refund actually firing, are the ones where the cost of catching a bad handoff clearly beats the cost of the extra hop. \nEverywhere else, you're probably just adding latency for the sake of feeling thorough.\n···\n## Final thoughts\nIf you're running a multi-agent pipeline today and want to try this without tearing the whole thing apart, don't start at the front of the chain. \nTrust me.\nStart at the last internal handoff before something external actually happens, right before an email sends, a record gets written, a decision gets acted on. That's the one boundary worth instrumenting first. \nGive it a week or two, see what the watchdog actually catches, and let that tell you whether it's worth pushing further back into the pipeline.\nYou don't need the full architecture on day one to get something out of this. \nNext time something breaks in a pipeline you own, ask yourself whether your evals would have caught it before the output itself looked wrong. \nMost of the time, if you're honest, the answer's no. That gap is exactly where the riskiest handoff in your system has been sitting the whole time, waiting for someone to look.\nYour final output can lie to you. The trajectory can't.\n···\n## Before you go!\nI write about the engineering decisions that decide whether an AI system holds up in production. You can [subscribe to my newsletter](https://thatcsguy.substack.com/?r=79lldp&utm_campaign=pub-share-checklist) if you'd like more of that.\n**Connect With Me**\n  * [LinkedIn](http://www.linkedin.com/in/nwekebenjamin)\n  * [Substack](https://thatcsguy.substack.com/)\n\n\n## Register Now to Continue Reading\n### Unlock Towards Data Science with a Free Account\nAll articles on Towards Data Science are available to registered users. Create a free account today to access this article and thousands more.\nRegister for freeSign in to your account\n* * *\n![](https://assets.insightmediagroup.io/media/1787657271062_x78p7k.jpeg)\nWritten By\n## [Benjamin Nweke](https://towardsdatascience.com/author/benjaminnweke/)\nBenjamin Nweke is a Computer Science graduate and technical writer exploring AI, machine learning, and data science. He writes about building reliable AI systems, model evaluation, agentic architectures, and the practical challenges that appear when machine learning moves beyond experiments. His work focuses on making technically complex ideas clear, useful, and grounded in real implementation.\n[See all from Benjamin Nweke](https://towardsdatascience.com/author/benjaminnweke/)\n[AI Agents](https://towardsdatascience.com/tag/ai-agents-2/)[Artificial Intelligence](https://towardsdatascience.com/tag/artificial-intelligence/)[Llm](https://towardsdatascience.com/tag/llm/)[Machine Learning](https://towardsdatascience.com/tag/machine-learning/)[Python](https://towardsdatascience.com/tag/python/)\nShare This Article\n  * [](https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Ftowardsdatascience.com%2Fwhy-most-multi-agent-systems-fail-even-when-evaluation-passes%2F&title=Why%20Most%20Multi-Agent%20Systems%20Fail%20Even%20When%20Evaluation%20Passes)\n  * [](https://www.linkedin.com/shareArticle?mini=true&url=https%3A%2F%2Ftowardsdatascience.com%2Fwhy-most-multi-agent-systems-fail-even-when-evaluation-passes%2F&title=Why%20Most%20Multi-Agent%20Systems%20Fail%20Even%20When%20Evaluation%20Passes)\n  * [](https://x.com/share?url=https%3A%2F%2Ftowardsdatascience.com%2Fwhy-most-multi-agent-systems-fail-even-when-evaluation-passes%2F&text=Why%20Most%20Multi-Agent%20Systems%20Fail%20Even%20When%20Evaluation%20Passes)\n\n\nTowards Data Science is a community publication. Submit your insights to reach our global audience and earn through the TDS Author Payment Program.\n[Write for TDS](https://towardsdatascience.com/submissions/)\n## Related Articles\n  * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2025/11/image-307.jpg)](https://towardsdatascience.com/how-to-scale-your-llm-usage/)\n## [How to Scale Your LLM Usage](https://towardsdatascience.com/how-to-scale-your-llm-usage/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nLearn how to increase LLM usage to achieve increased productivity\n[Eivind Kjosbakken](https://towardsdatascience.com/author/oieivind/)\nNovember 29, 20256 min read\n  * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2025/10/1_6zWBVRO6BeewCJYWKsvyyQ.jpg)](https://towardsdatascience.com/agentic-ai-from-first-principles-reflection/)\n## [Agentic AI from First Principles: Reflection](https://towardsdatascience.com/agentic-ai-from-first-principles-reflection/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nFrom theory to code: building feedback loops that improve LLM accuracy\n[Mariya Mansurova](https://towardsdatascience.com/author/miptgirl/)\nOctober 24, 202520 min read\n  * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2025/09/glenn-carstens-peters-0woyPEJQ7jc-unsplash-scaled-1.jpg)](https://towardsdatascience.com/building-video-game-recommender-systems-with-fastapi-postgresql-and-render-part-1/)\n## [Building Video Game Recommender Systems with FastAPI, PostgreSQL, and Render: Part 1](https://towardsdatascience.com/building-video-game-recommender-systems-with-fastapi-postgresql-and-render-part-1/)\n[Data Science](https://towardsdatascience.com/category/data-science/)\nDesigning a video game recommendations service with Steams API\n[Lucas See](https://towardsdatascience.com/author/seelcs12/)\nSeptember 25, 202521 min read\n  * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2020/06/0CAXefMdYi8CDyPc1-scaled.jpg)](https://towardsdatascience.com/automatic-notification-to-email-with-python-810fd357d89c/)\n## [Automatic Notification to Email with Python](https://towardsdatascience.com/automatic-notification-to-email-with-python-810fd357d89c/)\n[Machine Learning](https://towardsdatascience.com/category/artificial-intelligence/machine-learning/)\nAutomate your notification system with Python\n[Cornellius Wijaya](https://towardsdatascience.com/author/cornelliusyudhawijaya/)\nJune 13, 20204 min read\n  * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2026/08/ChatGPT-Image-Aug-2-2026-10_19_34-PM.jpg)](https://towardsdatascience.com/i-built-an-ai-data-agent-which-can-query-data-and-answer-business-questions-heres-how/)\n## [I Built an AI Data Agent Which Can Query Data and Answer Business Questions. Here’s How.](https://towardsdatascience.com/i-built-an-ai-data-agent-which-can-query-data-and-answer-business-questions-heres-how/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nA step-by-step guide to building a data agent and conversational interface that let business users to explore data in natural language without SQL\n[Jiayan Yin](https://towardsdatascience.com/author/jiayanyin.simba/)\nAugust 6, 202610 min read\n  * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2025/11/image-106.jpg)](https://towardsdatascience.com/how-to-automate-workflows-with-ai/)\n## [How to Automate Workflows with AI](https://towardsdatascience.com/how-to-automate-workflows-with-ai/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nLearn how to take a manual process and optimize it using AI\n[Eivind Kjosbakken](https://towardsdatascience.com/author/oieivind/)\nNovember 15, 20256 min read\n  * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2026/06/image-3.jpeg)](https://towardsdatascience.com/bi-is-dead-long-live-bi/)\n## [BI Is Dead, Long Live BI](https://towardsdatascience.com/bi-is-dead-long-live-bi/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nThe true bottleneck was never the analysis.\n[Mahdi Karabiben](https://towardsdatascience.com/author/mahdiqb/)\nJune 11, 20268 min read\n  * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2026/08/cover-image.png)](https://towardsdatascience.com/agentops-is-not-mlops-what-breaks-in-your-monitoring-stack-when-agents-go-to-production/)\n## [AgentOps Is Not MLOps: What Breaks in Your Monitoring Stack When Agents Go to Production](https://towardsdatascience.com/agentops-is-not-mlops-what-breaks-in-your-monitoring-stack-when-agents-go-to-production/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nThe five MLOps monitoring assumptions agents break, and which inherited signals now pass failed runs as healthy.\n[Mostafa Ibrahim](https://towardsdatascience.com/author/mostafaibrahim18/)\nAugust 31, 202610 min read\n  * [![](https://assets.insightmediagroup.io/media/wp-content/uploads/2025/10/1_6A4InWRTdVLXpqhUBuryBw.jpg)](https://towardsdatascience.com/deploy-an-openai-agent-builder-chatbot-to-a-website/)\n## [Deploy an OpenAI Agent Builder Chatbot to your Website](https://towardsdatascience.com/deploy-an-openai-agent-builder-chatbot-to-a-website/)\n[Agentic AI](https://towardsdatascience.com/category/artificial-intelligence/agentic-ai/)\nUsing OpenAI’s Agent Builder ChatKit\n[Thomas Reid](https://towardsdatascience.com/author/thomas_reid/)\nOctober 24, 202510 min read\n\n\n  * [YouTube](https://www.youtube.com/c/TowardsDataScience)\n  * [X](https://x.com/TDataScience)\n  * [LinkedIn](https://www.linkedin.com/company/towards-data-science/)\n  * [Threads](https://www.threads.net/@towardsdatascience)\n  * [Bluesky](https://bsky.app/profile/towardsdatascience.com)\n\n\n[](https://towardsdatascience.com/)\nYour home for data science and AI. The world's leading publication for data science, data analytics, data engineering, machine learning, and artificial intelligence professionals.\nSubscribe to Our Newsletter[Write For TDS](https://towardsdatascience.com/submissions/)•[About](https://towardsdatascience.com/about-towards-data-science-d691af11cc2f/)•[Advertise](https://contact.towardsdatascience.com/advertise-with-towards-data-science)•[Privacy Policy](https://towardsdatascience.com/privacy-policy/)•[Terms of Use](https://towardsdatascience.com/website-terms-of-use/)\n© Insight Media Group, LLC 2026•[The New Stack](https://thenewstack.io)•[roadmap.sh](https://roadmap.sh)•[Creator Network](https://insightmediagroup.io/creator-network)\n####  Recommended Articles \nClose\n  * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607686/image?etag=1782161464448&width=500&height=500) How Relevance Models Foreshadowed Transformers for NLP ](https://towardsdatascience.com/how-relevance-models-foreshadowed-transformers-for-nlp/ \"How Relevance Models Foreshadowed Transformers for NLP\")\n  * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607639/image?etag=1781250125848&width=500&height=500) Javascript Fatigue: HTMX Is All You Need to Build ChatGPT — Part 1 ](https://towardsdatascience.com/javascript-fatigue-you-dont-need-js-to-build-chatgpt/ \"Javascript Fatigue: HTMX Is All You Need to Build ChatGPT — Part 1\")\n  * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/607930/image?etag=1782124566054&width=500&height=500) The Subset Sum Problem Solved in Linear Time for Dense Enough Inputs ](https://towardsdatascience.com/subset-sum-problem-solved-in-linear-time-for-dense-enough-inputs/ \"The Subset Sum Problem Solved in Linear Time for Dense Enough Inputs\")\n  * [ ![](https://h030.towardsdatascience.com/rest/v2/contentStores/d7d3a4ab-7caf-4383-bbd2-fbf7fb4f1787/items/3660/image?etag=1780176246070&width=500&height=500) Does Your Company Have a Data Strategy? ](https://towardsdatascience.com/does-your-company-have-a-data-strategy-e8944dbec4ae/ \"Does Your Company Have a Data Strategy?\")\n\n\n",
  "body_base_path": "attachments/aa9c6400b486/pages/",
  "body_markdown_path": "attachments/aa9c6400b486/pages/archived-body.md",
  "local_html": "attachments/aa9c6400b486/pages/archived-body.html",
  "body_source": "prior_capture"
}