arxiv: 2604.21672 · v1 · submitted 2026-04-23 · 💰 econ.EM · q-fin.CP

Recognition: unknown

Agentic Artificial Intelligence in Finance: A Comprehensive Survey

Alex Ng, Chia-Yi Chien, Dora Xu, George Lin, Hanxi Xiao, Irene Aldridge, Jolie An, Kexin Deng, Michael Cao, Olivia Guo, Percy Lyu, Qi Wang, Riley Burke, Ruipeng Deng, Sheng Zhang, Shunran He, Sirui Zhang, Sirui Zhao, Waner Zheng, Weihang Lin, Xiaolong Zhao, Yichen Gao, Yihan Zhao, Yuanyuan Xue, Yun Zhang, Zheng Li

Pith reviewed 2026-05-08 12:59 UTC · model grok-4.3

classification 💰 econ.EM q-fin.CP

keywords agentic AIfinancial marketsautonomous systemsmarket efficiencysystemic riskregulatory complianceAI architecturerisk management

0 comments

The pith

Agentic AI systems autonomously reason and adapt in financial markets, delivering efficiency gains while creating fresh stability and oversight challenges.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper surveys recent work on agentic artificial intelligence applied to finance and argues that these systems mark a shift from earlier algorithmic tools. It claims the combination of goal-oriented autonomy, continuous learning, and multi-agent coordination lets the systems handle trading, liquidity, and risk tasks with little human input, which could raise overall market performance. At the same time the survey flags new downsides such as harder-to-predict market swings, compliance gaps, and opaque decisions that could affect the whole system. A reader would care because these agents are already moving from research into live trading desks and regulators need a map of both the upside and the exposure. The authors organize architectures, use cases, and governance ideas to show where further study can keep the benefits without amplifying the hazards.

Core claim

The survey establishes that agentic AI, set apart by its capacity for goal-oriented autonomy, continuous learning from experience, and coordination among multiple agents, is reshaping financial operations across system design, market uses, regulation, and systemic effects. The authors synthesize evidence that these traits enable stronger market efficiency, deeper liquidity provision, and more responsive risk management than prior algorithmic or generative approaches. They also document that the same traits generate distinct difficulties for maintaining stability, achieving regulatory compliance, ensuring interpretability of decisions, and containing systemic risk.

What carries the argument

Agentic AI distinguished by goal-oriented autonomy, continuous learning, and multi-agent coordination that operates with minimal human intervention in financial tasks.

If this is right

Agentic AI can raise market efficiency and improve liquidity provision through adaptive, goal-directed decisions.
Risk management becomes more dynamic because agents can learn continuously and coordinate across tasks.
Market stability faces new exposure from autonomous actions that may interact in unforeseen ways.
Regulatory compliance requires updated frameworks tailored to the interpretability and coordination features of these systems.
Systemic risk oversight must account for the reduced human intervention and the potential for coordinated agent behavior at scale.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Financial institutions may need to redesign oversight processes to monitor multi-agent coordination rather than single algorithms.
Regulators could use controlled test environments to measure real stability effects before wider deployment of agentic systems.
The same autonomy traits highlighted here could apply to other high-stakes domains where minimal human intervention is desired.
Quantitative studies could test whether the claimed efficiency gains scale linearly with the number of coordinated agents.

Load-bearing premise

The assumption that agentic AI is distinct from traditional algorithmic trading and generative AI through its specific capacities for goal-oriented autonomy, continuous learning, and multi-agent coordination, and that the surveyed literature fairly represents the current state of advances and implications.

What would settle it

A side-by-side comparison of deployed agentic AI systems and conventional algorithmic trading in live markets that finds no measurable difference in autonomy levels, learning behavior, coordination, efficiency outcomes, or stability impacts.

read the original abstract

The emergence of agentic artificial intelligence (AI) represents a fundamental transformation in financial markets, characterized by autonomous systems capable of reasoning, planning, and adaptive decision-making with minimal human intervention. This comprehensive survey synthesizes recent advances in agentic AI across multiple dimensions of financial operations, including system architecture, market applications, regulatory frameworks, and systemic implications. We examine how agentic AI differs from traditional algorithmic trading and generative AI through its capacity for goal-oriented autonomy, continuous learning, and multi-agent coordination. Our analysis shows that while agentic AI offers substantial potential for enhanced market efficiency, liquidity provision, and risk management, it also introduces novel challenges related to market stability, regulatory compliance, interpretability, and systemic risk. Through a systematic review of foundational research, technical architectures, market applications, and governance frameworks, this survey provides scholars and practitioners with a structured understanding of how agentic AI is reshaping financial markets and identifies critical research directions for ensuring that these systems enhance both operational efficiency and market resilience.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This is a broad survey on agentic AI in finance that sketches architectures, uses, and risks but provides no details on how the literature was chosen.

read the letter

The main thing here is that the paper is a literature survey claiming to be systematic, yet it gives no account of search strategy, databases, keywords, date range, or inclusion rules. That gap directly affects how much weight to put on its picture of efficiency gains versus stability and compliance problems. Without those details, the synthesis could be skewed or incomplete, and the claim that agentic AI is meaningfully different from algorithmic trading rests on unverified coverage of the prior work. The abstract asserts a balanced view of benefits and novel challenges, but readers cannot check whether the selected papers actually support that balance. What the paper does is organize the discussion into sections on system architecture, market applications, regulatory frameworks, and systemic implications. It pulls together existing ideas about goal-oriented autonomy, continuous learning, and multi-agent setups, and it flags practical issues like interpretability and market resilience. For someone new to the area, that structure offers a quick way to see the main threads without hunting through scattered sources. The soft spot is not minor because the central argument depends on the review being representative. A survey that calls itself comprehensive needs to show its work on selection; otherwise the positive and negative assessments remain assertions rather than demonstrated outcomes. This paper is mainly for readers who want an entry-level map of the topic, such as students, early-career researchers, or practitioners who need a high-level briefing before diving into primary papers. It adds little for specialists already following the cited literature. It deserves peer review because the topic is current and the outline is reasonable, but any referee should require a clear methods section and perhaps a table of included papers so the synthesis can be evaluated properly.

Referee Report

1 major / 1 minor

Summary. The paper is a comprehensive survey synthesizing recent advances in agentic AI across financial operations. It examines system architectures, market applications, regulatory frameworks, and systemic implications. The authors distinguish agentic AI from traditional algorithmic trading and generative AI via goal-oriented autonomy, continuous learning, and multi-agent coordination. They claim it offers substantial potential for enhanced market efficiency, liquidity provision, and risk management, while introducing novel challenges in market stability, regulatory compliance, interpretability, and systemic risk. The survey aims to provide scholars and practitioners with a structured understanding and to identify critical research directions for balancing efficiency and resilience.

Significance. If the underlying literature synthesis proves representative and balanced, the survey could be a useful organizing resource for an emerging interdisciplinary area, consolidating technical, application, and governance perspectives while highlighting open questions on deployment. The broad multi-dimensional coverage is a positive feature for readers seeking an entry point into the topic.

major comments (1)

[Abstract] Abstract: The text describes the work as 'a systematic review of foundational research, technical architectures, market applications, and governance frameworks,' yet supplies no details on search strategy, databases, keywords, date range, inclusion/exclusion criteria, or number of papers screened and included. Because the central claims about benefits versus novel risks rest on the completeness and balance of the selected evidence, this omission is load-bearing and prevents verification of the synthesis.

minor comments (1)

Clarify the precise definition and operational criteria used to classify a system as 'agentic' versus traditional algorithmic or generative AI, preferably with a short comparison table early in the manuscript.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive feedback on our manuscript. We address the major comment point by point below.

read point-by-point responses

Referee: [Abstract] Abstract: The text describes the work as 'a systematic review of foundational research, technical architectures, market applications, and governance frameworks,' yet supplies no details on search strategy, databases, keywords, date range, inclusion/exclusion criteria, or number of papers screened and included. Because the central claims about benefits versus novel risks rest on the completeness and balance of the selected evidence, this omission is load-bearing and prevents verification of the synthesis.

Authors: We agree that the abstract's reference to a 'systematic review' without accompanying methodological details is a valid concern, as it affects the ability to evaluate the balance of the synthesis. In the revised manuscript, we will add a dedicated 'Literature Search and Selection' subsection (likely in the Introduction or a new Methods section) that explicitly describes the search strategy. This will include: databases consulted (Google Scholar, arXiv, SSRN, Web of Science, and key journals in finance, economics, and AI); keywords and Boolean search strings (e.g., 'agentic AI' OR 'autonomous agents' AND 'finance' OR 'financial markets'); date range (primarily 2018–2024, with earlier foundational works included where relevant); inclusion/exclusion criteria (peer-reviewed articles, working papers, and reports focused on goal-oriented autonomy, multi-agent coordination, or applications in trading/risk/regulatory contexts; exclusion of purely theoretical AI papers without financial relevance); and approximate numbers screened and included. We will also note any limitations in coverage for this rapidly evolving field. This revision will strengthen transparency without altering the core claims. revision: yes

Circularity Check

0 steps flagged

No derivations or predictions present; survey is purely synthetic

full rationale

This paper is a literature survey with no equations, models, fitted parameters, or claimed first-principles derivations. Its central statements are descriptive syntheses of prior work rather than reductions of new results to the paper's own inputs. No load-bearing steps match any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a survey paper that introduces no new parameters, axioms, or entities and relies entirely on synthesis of existing published work.

pith-pipeline@v0.9.0 · 9098 in / 1027 out tokens · 105271 ms · 2026-05-08T12:59:46.824192+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages

[1]

Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges toward Responsible AI , shorttitle =

ge language models and reinforcement learning algorithms. This architecture provides a blueprint for building automated stress testing systems capa- ble of operating at various scales and complexity levels. The framework emphasizes world models—internal simulations agents use to envision outcomes without real-world experimentation—allowing agents to simul...

work page doi:10.1016/j.inffus.2019.12.012 2022
[2]

Bahrpeyma, F., and Reichelt, D. (2022). A review of the applications of multi-agent reinforcement learning in smart factories. Frontiers in Robotics and AI, 9, 1027340. Bao, W., and Liu, X. (2019). Multi Agent Deep Reinforcement Learning for Liquidation Strategy Analysis. arXiv preprint arXiv:1906.11046. Bongiorno, C., Manolakis, E., and Mantegna, R. N. (...

work page doi:10.1145/3580305.3599856 2022