Failure Modes & Fallbacks: What Agents Do When Things Go Wrong

{
  "byte_id": "agentic_dev_017",
  "title": "Failure Modes & Fallbacks: What Agents Do When Things Go Wrong",
  "level": "foundation",
  "domain": [
    "agentic-development",
    "reliability",
    "production"
  ],
  "intent": "Understand the most common ways agents fail in production and how to design explicit fallback strategies instead of silent breakdowns.",
  "core_idea": {
    "one_liner": "Agents will fail; the question is whether they fail safely.",
    "why_it_matters": [
      "Most agent failures are predictable.",
      "Silent failures destroy trust faster than explicit refusals.",
      "Fallbacks turn errors into controlled outcomes."
    ]
  },
  "definition": {
    "failure_mode": "A known way in which an agent can produce incorrect, incomplete, or unsafe behavior.",
    "fallback": "A predefined safe response or alternative path when a failure mode is detected."
  },
  "common_failure_modes": [
    {
      "mode": "Missing data",
      "symptom": "Agent answers confidently without evidence.",
      "fallback": "Ask for missing inputs or state that verification is not possible."
    },
    {
      "mode": "Tool failure",
      "symptom": "Timeouts, partial results, API errors.",
      "fallback": "Retry with backoff or switch to read-only / degraded mode."
    },
    {
      "mode": "Low confidence",
      "symptom": "Multiple conflicting answers retrieved.",
      "fallback": "Escalate to human-in-the-loop with options."
    },
    {
      "mode": "Guardrail violation",
      "symptom": "Request exceeds authority or scope.",
      "fallback": "Refuse and explain allowed alternatives."
    },
    {
      "mode": "Budget exhaustion",
      "symptom": "Agent hits cost or latency limits.",
      "fallback": "Return partial result with explanation."
    }
  ],
  "fallback_design_principles": [
    "Fallbacks must be explicit, not accidental.",
    "Fallbacks must be safe by default.",
    "Users must understand why a fallback happened."
  ],
  "micro_example": {
    "scenario": "Agent cannot retrieve required system metrics.",
    "bad_behavior": "Provides a generic guess.",
    "good_behavior": {
      "message": "I cannot verify the current system state because metrics are unavailable.",
      "options": [
        "Retry later",
        "Escalate to human support",
        "Provide a generic checklist instead"
      ]
    }
  },
  "failure_observability": [
    "Log failure type and trigger",
    "Count fallback frequency",
    "Alert if fallback rate spikes"
  ],
  "guards": [
    "Never answer critical questions without evidence.",
    "Fallback paths must be tested.",
    "Fallbacks must respect output contracts."
  ],
  "teach_it_in_english": {
    "simple_explanation": "A good agent knows when to stop instead of guessing.",
    "one_sentence_definition": "Fallbacks make agent failure predictable and safe."
  },
  "practical_checklist": [
    "What happens when data is missing?",
    "What happens when tools fail?",
    "Can the agent say 'I don’t know'?",
    "Are fallbacks visible and logged?"
  ],
  "tags": [
    "failure-modes",
    "fallbacks",
    "agent-reliability",
    "production"
  ],
  "meta": {
    "schema": "dkharlanau.dataset.byte",
    "schema_version": "1.1",
    "dataset": "agentic-bytes",
    "source_project": "cv-ai",
    "source_path": "agentic-bytes/agentic_dev_017.json",
    "generated_at_utc": "2026-02-03T14:33:32+00:00",
    "creator": {
      "name": "Dzmitryi Kharlanau",
      "role": "SAP Lead",
      "website": "https://dkharlanau.github.io",
      "linkedin": "https://www.linkedin.com/in/dkharlanau"
    },
    "attribution": {
      "attribution_required": true,
      "preferred_citation": "Dzmitryi Kharlanau (SAP Lead). Dataset bytes: https://dkharlanau.github.io"
    },
    "license": {
      "name": "",
      "spdx": "",
      "url": ""
    },
    "links": {
      "website": "https://dkharlanau.github.io",
      "linkedin": "https://www.linkedin.com/in/dkharlanau"
    },
    "contact": {
      "preferred": "linkedin",
      "linkedin": "https://www.linkedin.com/in/dkharlanau"
    },
    "canonical_url": "https://dkharlanau.github.io/datasets/agentic-bytes/agentic_dev_017.json",
    "created_at_utc": "2026-02-03T14:33:32+00:00",
    "updated_at_utc": "2026-02-03T15:29:02+00:00",
    "provenance": {
      "source_type": "chat_export_extraction",
      "note": "Extracted and curated by Dzmitryi Kharlanau; enriched for attribution and crawler indexing."
    },
    "entity_type": "agentic_byte",
    "entity_subtype": "level:foundation",
    "summary": "Understand the most common ways agents fail in production and how to design explicit fallback strategies instead of silent breakdowns."
  }
}