Orchestration · Software component
Web Navigation Agent
Software componentOrchestrationOrchestration & Toolsarc:WebNavigationAgent
An agent runtime that carries out multi-step web tasks by planning action sequences, selecting browser actions from the current page state, and tracking state across navigation steps until the user's goal state is reached.
Responsibility. Drives the per-task loop that chooses the next web interaction toward the user's goal.
Also known as: Web agent, Browser agent, Agentic commerce agent
Relationships
invokes dependency
is invoked by dependency
reads dependency
receives delegation from dependency
writes dependency
- Audit Log Store Ch3.3
- Conversation State Store abstract Ch3.3
- Working Memory Buffer abstract Ch3.3
emits telemetry to dynamic
receives data from dynamic
is constrained by control
is overridden by control
is audited by assurance
is evaluated by assurance
- LLM Judge abstract Ch3.3
- State Outcome Scorer Ch3.3
is monitored by assurance
Design guidance
- SHOULD generalize across site architectural patterns rather than executing memorized interaction sequences.
- SHOULD start production deployment with read-only information gathering before expanding to transactional operations.
- SHOULD maintain state and context across multi-turn conversations in which users supply requirements incrementally.
- SHOULD adapt when preferred items or paths are unavailable, e.g., by proposing alternatives matching user preferences.
Quantitative guidance
As stated by the sources; verify before use.
- Frontier agents reaching ~60% success on controlled WebArena drop to ~30% on live Online-Mind2Web sites (Ch3.3).
- MMInA: 1,050 real-world multimodal web tasks averaging 2.85 hops between sites and 12.9 actions per task (Ch3.3).
- Illustrative: an agent at 85% on e-commerce sites seen in development may reach only 40% on unfamiliar forum sites (Ch3.3).
Classification
- Patterns
- Multi-tool orchestration (web navigation plus auxiliary tools)Staged rollout: read-only before transactionalMulti-turn conversational refinement
- Technologies
- Ask InstacartAmazon Buy for MeSkyvern
- Quality attributes
- Reliability (ISO/IEC 25010 | NIST AI RMF: valid and reliable)Functional suitability: correctness and validity (ISO/IEC 25010 | NIST AI RMF: valid)Performance efficiency (ISO/IEC 25010)Cost efficiency
- Risks mitigated
- Brittleness to unfamiliar site architecturesDynamic website state (items out of stock mid-session)
Sources
- Ch3.3: T. Nguyen, "Web Navigation and Interaction Benchmarks," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 3.3. ISBN: 9798244538229.