arXiv:2608.18110 catalogs agentic AI architectures and adoption factors from 2023-2026 literature
arXiv:2608.18110 aggregates recent agentic AI research into architecture, applications, and adoption. It identifies gaps in evaluation standards and proposes a quality-mediated adoption model. The review supplies researchers with a structured map of current systems and open questions.
The paper traces agency from early symbolic planners through ReAct, Auto-GPT, and 2025 multi-agent deployments. It extracts working principles from 12 application domains and maps them to measurable quality attributes including autonomy, transparency, and error recovery. Data tables list benchmark scores from GAIA, WebArena, and ToolBench runs reported in the cited works.
Adoption analysis draws on TAM extensions and identifies latency, cost per task, and auditability as dominant predictors. The proposed framework adds system quality as a mediator between perceived usefulness and intention to use, supported by path coefficients from prior ERP and RPA studies. Gaps noted include absence of standardized failure taxonomies and limited longitudinal deployment data.
Operational implications center on integration cost: organizations must instrument telemetry for each agent loop and maintain human override thresholds below 5 percent intervention rate. Future work sections flag needs for verifiable execution traces and cross-agent protocol benchmarks. No production deployment metrics beyond 2025 pilots appear in the reviewed corpus.
Haque et al.: GAIA agent success rate exceeds 65 percent on 2026 tasks with full telemetry by Q4 2027
Sources (3)
- [1]Primary Source(https://arxiv.org/abs/2608.18110)
- [2]Supporting Source(https://arxiv.org/abs/2210.03629)
- [3]Supporting Source(https://arxiv.org/abs/2303.17651)