{
  "schemaVersion": 3,
  "id": "article:long-running-sessions",
  "slug": "long-running-sessions",
  "title": "Long-Running Sessions: Keeping AI Work Coherent Over Time",
  "canonicalPath": "/articles/long-running-sessions/",
  "sourcePath": "content/articles/2026/long-running-sessions/article.md",
  "agentBriefPath": "content/articles/2026/long-running-sessions/agent.md",
  "thesis": "Long-running AI sessions need clear goals, summaries, memory, checkpoints, context pruning, and stopping rules to avoid drift and wasted work.",
  "status": "published",
  "maturity": "seed",
  "publishedAt": "2026-06-29",
  "updatedAt": "2026-06-29",
  "audiences": [
    "general",
    "students",
    "builders"
  ],
  "topics": [
    "ai-agents",
    "ai-literacy"
  ],
  "series": {
    "slug": "ai-demystified",
    "title": "AI, De-Mystified",
    "order": 11,
    "role": "chapter"
  },
  "claims": [
    {
      "id": "claim-001",
      "claim": "A long-running session is useful only when the system can remember what matters, recognize progress, and decide when to stop.",
      "confidence": "high",
      "status": "core",
      "evidence": [
        {
          "sourceId": "source-generative-agents",
          "snippet": "Generative agents use a memory stream, retrieval, and reflection to maintain coherent behavior over extended simulated time.",
          "supports": "direct",
          "assessedAt": "2026-06-29"
        }
      ],
      "counterevidence": [
        {
          "summary": "Memory and summarization are selective. Important details can be compressed out or retrieved incorrectly, so remembering what matters is not guaranteed.",
          "assessedAt": "2026-06-29"
        }
      ]
    },
    {
      "id": "claim-002",
      "claim": "Keeping extended work coherent is already familiar from workflow orchestration, durable execution, project management, and process control.",
      "confidence": "high",
      "status": "landscape",
      "evidence": [
        {
          "sourceId": "source-react-paper",
          "snippet": "Reasoning and acting have a long history in AI; ReAct explicitly connects the loop of thought, action, and observation to prior work in reinforcement learning and decision making.",
          "supports": "background",
          "assessedAt": "2026-06-29"
        }
      ],
      "counterevidence": [
        {
          "summary": "Older fields usually assume formally defined goals, stable environments, or fixed stages. LLM-based sessions face ambiguous, shifting goals and improvised steps.",
          "assessedAt": "2026-06-29"
        }
      ]
    },
    {
      "id": "claim-003",
      "claim": "In practice, long-running sessions combine summarization, checkpoints, context pruning, and prompt caching to keep the active window focused without losing the goal.",
      "confidence": "medium-high",
      "status": "design",
      "evidence": [
        {
          "sourceId": "source-self-refine-paper",
          "snippet": "Self-Refine iteratively refines outputs using self-generated feedback, showing how a loop of generation and summary can extend a single session.",
          "supports": "direct",
          "assessedAt": "2026-06-29"
        },
        {
          "sourceId": "source-react-paper",
          "snippet": "ReAct interleaves reasoning traces and actions, relying on the session history to stay focused on the task while observing new information.",
          "supports": "direct",
          "assessedAt": "2026-06-29"
        },
        {
          "sourceId": "source-lumer-prompt-caching-agents",
          "snippet": "Prompt caching reduces the cost and latency of repeated context in multi-turn sessions, making it more practical to keep long-running sessions coherent.",
          "supports": "direct",
          "assessedAt": "2026-06-29"
        }
      ],
      "counterevidence": [
        {
          "summary": "Not every useful task needs a long session; some problems are best solved by a single, carefully crafted prompt or a stateless function.",
          "assessedAt": "2026-06-29"
        }
      ]
    },
    {
      "id": "claim-004",
      "claim": "Without summaries, checkpoints, and stopping rules, long-running sessions drift, waste resources, or resume in broken states.",
      "confidence": "medium",
      "status": "risk",
      "evidence": [
        {
          "sourceId": "source-reflexion-paper",
          "snippet": "Reflexion shows that language agents can benefit from verbal reinforcement signals to correct trajectory, implying that unguided trajectories can drift or repeat errors.",
          "supports": "indirect",
          "assessedAt": "2026-06-29"
        },
        {
          "sourceId": "source-generative-agents",
          "snippet": "Long-term memory retrieval in generative agents can fail to surface the most relevant context, leading to behavior that diverges from prior decisions.",
          "supports": "indirect",
          "assessedAt": "2026-06-29"
        }
      ],
      "counterevidence": [
        {
          "summary": "Simple exit conditions such as cost or time budgets can prevent runaway work, but they do not guarantee the original goal has been reached.",
          "assessedAt": "2026-06-29"
        }
      ]
    },
    {
      "id": "claim-005",
      "claim": "Prompt caching can cut the cost and latency of repeated context in long sessions, making extended sessions more practical.",
      "confidence": "medium-high",
      "status": "design",
      "evidence": [
        {
          "sourceId": "source-lumer-prompt-caching-agents",
          "snippet": "A cross-provider evaluation of multi-turn research agents with 10,000-token system prompts found prompt caching reduced API costs by 41–80% and improved time-to-first-token by 13–31%.",
          "supports": "direct",
          "assessedAt": "2026-06-29"
        }
      ],
      "counterevidence": [
        {
          "summary": "Caching benefits depend on context size, provider implementation, and how much of the prompt is repeated; small or highly varied sessions may see little gain.",
          "assessedAt": "2026-06-29"
        }
      ]
    }
  ],
  "sources": [
    {
      "id": "source-react-paper",
      "title": "ReAct: Synergizing Reasoning and Acting in Language Models",
      "url": "https://arxiv.org/abs/2210.03629",
      "type": "paper",
      "accessed": "2026-06-29"
    },
    {
      "id": "source-self-refine-paper",
      "title": "Self-Refine: Iterative Refinement with Self-Feedback",
      "url": "https://arxiv.org/abs/2303.17651",
      "type": "paper",
      "accessed": "2026-06-29"
    },
    {
      "id": "source-generative-agents",
      "title": "Generative Agents: Interactive Simulacra of Human Behavior",
      "url": "https://arxiv.org/abs/2304.03442",
      "type": "paper",
      "accessed": "2026-06-29"
    },
    {
      "id": "source-reflexion-paper",
      "title": "Reflexion: Language Agents with Verbal Reinforcement Learning",
      "url": "https://arxiv.org/abs/2303.11366",
      "type": "paper",
      "accessed": "2026-06-29"
    },
    {
      "id": "source-lumer-prompt-caching-agents",
      "title": "Cross-Provider Evaluation of Prompt Caching in Multi-Turn Research Agents",
      "url": "https://arxiv.org/abs/2601.06007v2",
      "type": "paper",
      "accessed": "2026-06-29"
    }
  ],
  "related": [
    {
      "type": "article",
      "id": "article:loops-vs-goals"
    },
    {
      "type": "topic",
      "id": "topic:ai-agents"
    }
  ],
  "agentInstructions": [
    "Use claim IDs as the retrieval unit.",
    "Treat maturity=seed as an explicit uncertainty marker.",
    "Do not present long-running AI sessions as capable of remembering everything or running unattended indefinitely.",
    "When summarizing, preserve the plain-language-first, technical-depth-later structure."
  ],
  "provenance": {
    "createdAt": "2026-06-29",
    "createdBy": "human",
    "agents": [
      {
        "role": "drafting",
        "model": "kimi",
        "invokedAt": "2026-06-29",
        "inputHash": "sha256:0000000000000000000000000000000000000000000000000000000000000000",
        "outputHash": "sha256:eb8ff1fe9565aae89289fea8c2dd4c377e5dc02960e9f1f82e63fb0f6dccc6fd"
      },
      {
        "role": "review",
        "model": "kimi",
        "invokedAt": "2026-06-29",
        "inputHash": "sha256:0000000000000000000000000000000000000000000000000000000000000000",
        "outputHash": "sha256:eb8ff1fe9565aae89289fea8c2dd4c377e5dc02960e9f1f82e63fb0f6dccc6fd"
      }
    ],
    "reviews": [
      {
        "reviewer": "agent",
        "reviewedAt": "2026-06-29",
        "status": "approved",
        "scope": [
          "claims",
          "tone",
          "privacy",
          "scope"
        ],
        "notes": "Sibling-agent review against article structure. Privacy scan passed. No proprietary or personal content detected.",
        "contentHash": "eb8ff1fe9565aae89289fea8c2dd4c377e5dc02960e9f1f82e63fb0f6dccc6fd"
      },
      {
        "reviewer": "human",
        "reviewedAt": "2026-06-29",
        "status": "approved",
        "scope": [
          "thesis",
          "examples",
          "tone",
          "safety"
        ],
        "notes": "Human author approved the draft for publication.",
        "contentHash": "eb8ff1fe9565aae89289fea8c2dd4c377e5dc02960e9f1f82e63fb0f6dccc6fd"
      }
    ],
    "policy": {
      "id": "policy:default",
      "version": "1.0.0"
    }
  },
  "contentHash": "eb8ff1fe9565aae89289fea8c2dd4c377e5dc02960e9f1f82e63fb0f6dccc6fd",
  "generatedAt": "2026-06-29T00:00:00.000Z",
  "articleUrl": "https://aura-knowledge.github.io/articles/long-running-sessions/",
  "agentJsonPath": "/agents/articles/long-running-sessions.json",
  "agentMarkdownPath": "/agents/articles/long-running-sessions.md",
  "sourceRepoPath": "content/articles/2026/long-running-sessions/article.md",
  "sourceGitHubUrl": "https://github.com/aura-knowledge/aura-knowledge.github.io/blob/main/content/articles/2026/long-running-sessions/article.md",
  "tokenEstimate": 496,
  "sectionOutline": [
    {
      "id": "plain-english-meaning",
      "title": "Plain English Meaning"
    },
    {
      "id": "existing-concept-it-resembles",
      "title": "Existing Concept It Resembles"
    },
    {
      "id": "what-is-actually-new",
      "title": "What Is Actually New?"
    },
    {
      "id": "how-it-works-in-practice",
      "title": "How It Works In Practice"
    },
    {
      "id": "where-it-helps",
      "title": "Where It Helps"
    },
    {
      "id": "where-it-fails",
      "title": "Where It Fails"
    },
    {
      "id": "academic-connections",
      "title": "Academic Connections"
    },
    {
      "id": "practical-checklist",
      "title": "Practical Checklist"
    },
    {
      "id": "the-de-hype-check",
      "title": "The De-Hype Check"
    },
    {
      "id": "open-questions",
      "title": "Open Questions"
    }
  ]
}