Loading article...
Loading article...
Stream/Area:
-
Subject:
-
Paper ID:
IJSREM66757
Volume:
10 |
8 |
Aug 2026
Pages:
1-4
Publication Frequency:
Monthly (12 issues per year)
License:International Journal of Scientific Research in Engineering and Management follows the Creative Commons Attribution 4.0 (CC BY 4.0) License, allowing free use and distribution with proper attribution.
Independent AI Researcher
Software Engineer
Independent AI Researcher
Software Engineer
AI agents increasingly interact with application programming interfaces, enterprise systems, databases, workflow engines, and tool servers that can change business state. This creates a reliability problem that is different from ordinary response correctness: a failed response is not always a failed operation, and a successful response is not always a successful operation. This paper proposes a failure recovery engineering framework for agentic systems that perform side-effecting tool operations. The framework formalizes a failure taxonomy, introduces an Action Ledger pattern for durable business-operation memory, defines idempotency as a reasoning boundary for autonomous tool use, and proposes an explicit recovery state machine for unknown-state conditions. It also introduces a conceptual Failure Recovery Score (FRS) for evaluating classification, retry decisions, duplicate prevention, state verification, escalation correctness, and recovery outcome. The paper positions Microsoft Foundry as one possible implementation environment for agent orchestration, tool integration, tracing, and evaluation, while clearly distinguishing platform capabilities from application-owned reliability semantics. The contribution is a practical systems-engineering model for safer production AI agents.
KEYWORDS
Production AI agents are not made reliable by improving model intelligence alone. They become reliable when engineers define explicit boundaries around uncertainty, tool side effects, durable state, recovery decisions, and human intervention. The paper contributes a practical framework for failure recovery engineering in agentic systems: a failure taxonomy, richer tool contracts, the Action Ledger pattern, idempotency as a reasoning boundary, an explicit recovery state machine, and the Failure Recovery Score. Together, these concepts provide a foundation for safer production agents that act reliably even when the system cannot immediately prove what happened.
1. Microsoft Learn, "What is Microsoft Foundry Agent Service?" https://learn.microsoft.com/en-us/azure/foundry/agents/overview 2. Microsoft Learn, "Agent tracing overview - Microsoft Foundry." https://learn.microsoft.com/en-us/azure/foundry/observability/concepts/trace-agent-concept 3. Microsoft Learn, "Evaluate your AI agents - Microsoft Foundry." https://learn.microsoft.com/en-us/azure/foundry/observability/how-to/evaluate-agent 4. T. Pan, "Agent Idempotency: Why Your AI Agent Sends That Email Twice," 2026. https://tianpan.co/blog/2026-04-10-agent-idempotency-making-ai-side-effects-safely-retryable 5. T. Garg, "Building Reliable AI Agents: Patterns for Failure Recovery, Observability, and Safe Autonomy," 2026. https://tanujgarg.com/blog/ai-agent-reliability-patterns 6. OpenLegion, "AI Agent Reliability: Circuit Breakers, Backoff, and Dead-Letter Queues," 2026. https://www.openlegion.ai/en/learn/ai-agent-reliability 7. Codelit.io, "Idempotency in Distributed Systems: Retry Safety, Deduplication & Patterns," 2026. https://codelit.io/blog/idempotency-distributed-systems 8. Databricks, "What is AI Agent Evaluation?" https://www.databricks.com/blog/what-is-agent-evaluation 9. A. K. Padhy, "Evaluating AI Agents in Practice: Benchmarks, Frameworks, and Lessons Learned," InfoQ, 2026. https://www.infoq.com/articles/evaluating-ai-agents-lessons-learned/ 10. B. Xu, "AI Agent Systems: Architectures, Applications, and Evaluation," arXiv:2601.01743, 2026. https://arxiv.org/html/2601.01743v1
Publication Date
8/14/2026
Peer Review Type
Double-Blind Peer Review
Review Rounds: 2
Documents
Article Views
2
Total views since publication
PDF Downloads
6
Total downloads since publication
Geographic Reach
5
Digital Preservation: This article is permanently archived on Zenodo and complies with all data preservation standards across major indexing platforms.
Plagiarism Screening
This manuscript has been checked for plagiarism using TurnitinPlagiarismChecker software.
Conflict of Interest
The authors declare no conflict of interest related to this research work.
Funding Information
No funding information available for this research.
Related research in the same subject area
Loading similar articles...