Evidano home
Published in Electronic Commerce Research and ApplicationsResearch SpotlightUniversity of Gothenburg

Research Spotlight: When AI Moves From Prediction to Coordination in Quick Commerce

Evidano7 min read
Shahryar Sorooshian, Professor of Business Administration at the University of Gothenburg

A customer taps an order button and a clock starts. In the minutes that follow, stock has to be located, picking sequenced, a courier positioned, a route adjusted, and the customer kept informed while demand, traffic, and rider availability continue to change. That dense web of decisions makes quick commerce a revealing setting for a larger question in artificial intelligence: what happens when AI does not merely predict or recommend, but begins to coordinate actions across a system?

In a new review in Electronic Commerce Research and Applications, Shahryar Sorooshian, Navidreza Ahadi, and Pouria Liravi examine how research on multi-agent systems, reinforcement learning, simulation, market mechanisms, and bounded large-language-model decision support is converging around the idea of agentic AI in rapid fulfilment. Their central argument is deliberately cautious: the field shows substantial progress in autonomous logistics coordination, but not yet mature and widespread agentic AI deployment in live quick-commerce operations.

We invited one of the authors, Shahryar Sorooshian of the University of Gothenburg, to share more about how the field can move from technical capability to responsible use.

About the article

Research question

How does prior work on autonomous coordination, including multi-agent systems, reinforcement learning, simulation, auctions, and bounded large-language-model decision support, inform the emerging concept of agentic AI in quick-commerce fulfilment? The review also asks what the literature reveals about technical performance, human oversight, governance, labor, customers, and sustainability.

Key finding

Across 16 eligible studies, technical and operational concerns dominate, while governance and other sociotechnical concerns remain comparatively underdeveloped. The authors therefore argue that agentic AI in quick commerce is currently better understood as an emerging conceptual extension of autonomous logistics coordination than as a mature operational reality.

Why this research matters

Quick commerce compresses many interdependent decisions into very short time windows. Demand can shift suddenly, courier availability can change, traffic can disrupt routes, and a decision about one order can affect the capacity available for the next. This makes the sector an unusually clear test case for systems that can adapt, coordinate, and act over time rather than simply produce a prediction.

The review also matters because efficiency is only one part of the story. If autonomous systems influence dispatch, incentives, substitutions, service recovery, or customer communication, then questions about worker autonomy, fairness, explainability, privacy, accountability, and trust become part of system performance. The paper captures this tension plainly: “The overall picture, nonetheless, is one of technical capability outpacing governance.”

Professor Sorooshian’s perspective connects this literature to its wider significance: “In my view, the most important implication of this study is that research on multi-agent systems and agent coordination for Q-commerce has advanced considerably faster than research on governance frameworks and socio-technical safeguards. While technical and simulation modeling studies remain important, I encourage researchers to increasingly incorporate empirical investigations into their work, particularly those addressing labor impacts, accountability, privacy, and equity, to ensure more comprehensive and balanced coverage of the field.”

Key findings from the article

Three contributions stand out from the review. Together, they show both how far autonomous coordination research has progressed and where the evidence remains thin.

  • Technical coordination is the center of the field. The keyword maps place multiagent systems, reinforcement learning, logistics, last-mile delivery, decision making, optimization, and real-time systems at the core of the literature. Across the 16 studies, the authors identify a progression from classical operational research and decentralized logistics toward learning-based and more agentic forms of coordination.
  • The evidence for mature agentic AI deployment is still limited. Much of the reviewed work relies on modelling, simulation, and computational experiments. Studies demonstrate potential gains in routing, courier balancing, delivery zone planning, trading, and fleet use, but the paper repeatedly cautions that these results should not be treated as proof of widespread real-world deployment or universal superiority over centralized approaches.
  • Governance and human consequences lag behind technical development. Labor, ethics, transparency, privacy, regulation, and accountability are weak or absent in the bibliometric maps, and the sociotechnical stream is much smaller than the operational one. The authors call for more evidence on worker experience, customer trust, fairness, auditability, environmental effects, and the conditions under which autonomous coordination should be used at all.

Professor Sorooshian reflected on what surprised him most: “The literature on Q-commerce reveals a striking absence of a few themes, such as labor conditions, transparency, and oversight, in contrast to the heavy focus on technical aspects like multi-agent reinforcement learning and simulations. This gap is surprising given the high-stakes, labor-intensive nature of Q-commerce, where autonomous systems can affect work and urban environments. The disparity between the advancement of technical capabilities and the readiness for governance presents a significant concern.”

Methods and research approach

The study used a structured conceptual review design. The authors searched Scopus in April 2026 using terms covering agentic or autonomous AI and quick-commerce-related rapid fulfilment, then screened titles, abstracts, and full texts using a PRISMA-guided process. Of 29 records initially identified, 16 studies were retained for the final corpus.

The analysis combined bibliometric mapping in VOSviewer with qualitative thematic synthesis of the full texts. The bibliometric component examined keyword co-occurrence, cluster structure, and temporal patterns, while the thematic synthesis used a deductive and inductive coding process to interpret operational, market, sustainability, and governance themes. Evidano, then known as AILYZE, supported preliminary organization and coding outputs, while the authors manually reviewed the full texts, compared outputs with the bibliometric results, refined the codes, and determined the final themes and interpretations.

How Evidano supported the research

The paper reports that Evidano, then known as AILYZE, supported theme coding, pattern identification, organization of text segments, document summaries, possible codes, and provisional thematic groupings. In the authors’ words, “AILYZE was used to support the generation and organization of preliminary analytical outputs,” and they immediately clarify that “These outputs were not treated as final findings.”

Human review remained central to the workflow. The authors manually reviewed the full texts, compared the software-generated outputs with the bibliometric results, refined and reorganized the codes, and developed the final theme labels and interpretations reported in the article. In other words, AI assisted the organization and preliminary synthesis of evidence, while scholarly judgment and responsibility remained with the research team.

Professor Sorooshian said: “AILYZE helped the authors save some time by facilitating certain aspects of their efforts to summarize, thematically group, and organize the literature.”

Broader implications

For researchers, the review points to a clear next phase. The field needs more evidence from live settings, more qualitative and mixed-methods work with workers, managers, and customers, and broader geographic coverage across different labor markets, infrastructures, and regulatory environments. This would help test whether the advantages seen in simulations persist when autonomous systems meet the complexity of real organizations and cities.

For managers and policymakers, the article argues for governance before scale. Decisions about what a system may do independently, when a person must intervene, how exceptions are escalated, how decisions are recorded, and how workers and customers can understand or challenge automated outcomes should be designed alongside performance goals. The review also suggests that speed and cost should not be the only measures of success, with fairness, trust, complaints, service recovery, and environmental impact also requiring attention.

Asked for advice, Professor Sorooshian said: “Build the rules and the human checks in from day one. Speed and optimisation alone are never enough reason to switch something on. In fast-moving settings like quick commerce, these systems can improve routing, dispatching, and coordination. At the same time, they may influence how people work, what incentives they face, how customers experience the service, and who is accountable when something goes wrong. Right now, technical developments are advancing faster than the frameworks needed for fairness, transparency, and meaningful human oversight. Without appropriate safeguards in place, efficiency gains may also introduce new challenges.”

The lasting contribution of this review is not a claim that agentic AI has already transformed quick commerce. It is a more useful argument: that rapid fulfilment exposes the conditions under which autonomous coordination becomes consequential, and that the next stage of research must evaluate technical performance together with accountability, human experience, and public value.

About Shahryar Sorooshian

Shahryar Sorooshian, Professor of Business Administration at the University of Gothenburg

Shahryar Sorooshian is a Professor in the Department of Business Administration at the University of Gothenburg. His work focuses on industrial management, business engineering, and management engineering, with particular emphasis on operations and supply chains and the interface of technology, sustainability, and business management. He teaches supply chain management, operations management, information systems, and management science, and he supervises research across academic levels. His recent research includes quick commerce and last mile delivery, which connects directly to this review of agentic AI and autonomous coordination in rapid fulfilment.

About the authors

  • Shahryar Sorooshian (Department of Business Administration, University of Gothenburg, Sweden, and Faculty of Engineering and Sustainable Development, University of Gävle, Sweden)
  • Navidreza Ahadi (Derby International Business School, University of Derby, UK)
  • Pouria Liravi (Derby International Business School, University of Derby, UK)

Read the article

Read the full research article, “From multi-agent logistics to agentic artificial intelligence: review of autonomous coordination in Q-commerce-related last-mile fulfilment,” published in Electronic Commerce Research and Applications in 2026.

Full citation

Sorooshian, S., Ahadi, N., & Liravi, P. (2026). From multi-agent logistics to agentic artificial intelligence: Review of autonomous coordination in Q-commerce-related last-mile fulfilment. Electronic Commerce Research and Applications, 79, 101631. doi.org/10.1016/j.elerap.2026.101631

View the publication

We are grateful to the authors for their contribution to research on agentic AI and quick commerce, and for referencing Evidano, then known as AILYZE, as part of their research process.

Other published studies and evaluations that used Evidano on comparable material.