E-Commerce Leaders Point to Infrastructure Gaps Limiting Autonomous AI Operations

Despite a surge in artificial intelligence software targeting online merchants, e-commerce agency executives report that current operational tools remain largely reactive, flagging account errors while leaving human managers to manually apply fixes. The operational limitations of current store management platforms and the infrastructure required for autonomous store operations were detailed in interviews with seven industry leaders, as reported by The Next Web (https://thenextweb.com/news/ai-commerce-gap-ecommerce-leaders-ai-agents).

According to Cyril Golub, founder and chief executive of operations layer developer Jinnify.ai, modern e-commerce agencies are overburdened by software proliferation rather than a lack of tooling. Golub, an angel investor who previously co-founded e-commerce software maker Aheadworks before leading it to a 2019 exit, observed that every additional SaaS dashboard or AI copilot adds another interface to learn, interpret, and manually connect to core business outcomes.

Agency operators noted that the most time-consuming operational task is backend catalog maintenance rather than marketing or pay-per-click optimization. Sam Shah, founder of e-commerce agency Desverto, explained that store teams spend substantial time resolving suppressed ASIN listings, broken parent-child variations, catalog overwrites, 8541 system errors, and hazmat compliance documentation. Shah noted that while specialized detection tools like Data Dive spot operational issues, resolving them still requires human managers to file and follow up on Seller Support and Brand Registry cases.

Steven Pope, founder of US agency My Amazon Guy, which manages 450 client brands, stated that current SaaS products fail to reliably deliver end-to-end catalog resolution. Pope detailed how catalog maintenance involves continuously navigating flat files, category templates, inventory feeds, and shifting marketplace rules. He emphasized that the primary bottleneck remains exception management across thousands of minor catalog issues that require context and judgment to edit safely without damaging retail readiness, search indexing, or account compliance.

Current AI products also struggle to align tactical recommendations with overall business strategy and account lifecycles. Adnan Aslam, chief executive of UK-based agency Sellonics, noted that existing software lacks a unified control layer connecting specific campaign changes to broader commercial goals. Aslam explained that effective automation requires systems to concurrently evaluate advertising budgets, profit margins, inventory levels, and competitor rankings, noting that standard AI models cannot account for the realistic constraints of newly launched brand accounts seeking to rank against established sellers.

Connecting large language models directly to Amazon's Selling Partner API has similarly failed to yield reliable autonomous account managers due to data noise and hallucinated recommendations. Klaidas Siuipys, founder and chief executive of Vilnius, Lithuania-based agency AMZ Bees, noted that while protocols like Model Context Protocol enable systems to pull raw account data, generic models lack the contextual filtering needed to distinguish critical account threats from minor log events. Golub added that using general-purpose frontier models for raw commerce data processing requires deterministic pre-processing software, persistent strategic memory, operational execution guardrails, and automated feedback loops.

Industry executives contend that software must evolve from decision-support recommendations to direct system execution. Antons Sapriko, chief executive of European software agency Scandiweb, stated that tools requiring merchants to manually copy recommendations into seller portals fail to fundamentally change operating models. Ruslan Fazlyev, founder of e-commerce platform Ecwid, which reached a $500 million exit, remarked that after two decades of simplifying how merchants build online storefronts, the industry's next focus is building infrastructure that simplifies how merchants operate them.

To address these limitations, operators advocate for catalog control systems built around confidence-based automation. Under this proposed framework, low-risk operational edits execute automatically, medium-risk updates enter approval queues, and high-risk changes require human authorization accompanied by impact estimates. Golub, who previously outlined this paradigm on Claus Lauter's podcast The Ecommerce Coffee Break, stated that AI should function as a specialized assistant operating within explicit boundaries, laying the groundwork for agentic systems that manage seller workflows as automated agents increasingly shape consumer shopping decisions.