The inaugural year of supply chain AI agents begins here: Deep Insights unveils 46 specialized AI agents.

A medium‑scale supply chain disruption typically results in losses equivalent to 1%–4.5% of a company’s annual revenue. By 2026, 40% of enterprise applications will incorporate task‑oriented AI agents. — Gartner Over the past year, “agent” has been one of the most frequently cited terms in the supply chain industry. However, relatively few organizations have successfully moved agents beyond PowerPoint presentations and demos to operational deployment—into customers’ warehouses, vehicle fleets, and customs brokerage operations.

At Deep Insights, we’ve thoroughly reorganized the industry know-how we’ve accumulated over the past 22 years, along with our 55+ industry modules, 90+ business scenarios, and 46 specialized AI agents that are now being deployed at scale to serve our customers. 
 

If you’re also wondering how AI agents can truly integrate into your supply chain operations, you’re welcome. Click here to schedule a product demo · Schedule a free AI Agent consultation for shipping rates or warehouse diagnostics.

The following is an excerpt from CEO Liu Bin’s presentation at the ACE2026 16th Annual Supply Chain Innovation Summit, which will provide a clear explanation of how we define and understand “Agent.”

Supply chain execution—why is it stuck here?

Before discussing Agents, let’s first address the issue at hand. Today, the operational layer of China’s supply chain is simultaneously constrained by four factors: 
First, the human resources bottleneck. Warehouse workers are hard to recruit, and scheduling staff are difficult to retain; labor costs rise by more than 10% annually, while expertise cannot be quickly replicated—when a seasoned scheduler leaves, operational disruptions within six months are almost inevitable. 
Second, data silos. Multiple systems—WMS, TMS, ERP, and OA—operate in parallel, with data manually cross-checked repeatedly in Excel. If a single node fails, the entire workflow is impacted. 
Third, excessively passive. It’s the norm for most supply chain organizations that issues aren’t detected until after they’ve already occurred. Yet every hour of delay translates into both a loss in customer experience and tangible costs. 
Fourth, optimize bottlenecks. Even the most experienced professionals have their own cognitive limits. Faced with complex, massive datasets, human eyes simply cannot identify all the opportunities for cost reduction. 
 

These four tasks happen to correspond to the four types of problems that AI Agents excel at addressing: repetitive work, cross-system collaboration, real-time monitoring, and pattern recognition in massive datasets. This is also the fundamental reason why we have chosen AI Agents as the core product paradigm for the next decade.

Our Understanding of Agents (I): Four Fundamental Issues in Technical Implementation

Market narratives around Agents are often conflated. After internal clarification, we’ve broken down “technical implementation” into four distinct yet mutually reinforcing questions. Any team looking to build an Agent must first determine which type they’re pursuing.

01 Human-Used Agent —— Human-AI Collaboration

This is the type of task that is easiest to understand today and also the most straightforward to implement as a Copilot. At its core, it operates on a human-in-the-loop model: the agent serves as an expert assistant, freeing humans from low-value, highly repetitive tasks, while the ultimate authority over decision-making, confirmation, and approval remains with the human.

In the supply chain, typical agent roles include: when scheduling vehicles, the agent provides an optimal load‑matching plan; when handling price inquiries, the agent offers recommended quotes and risk alerts; and when conducting analyses, the agent generates visualizations and actionable insights.

It addresses the issue of “efficiency,” with a very clear boundary—humans remain the primary agents of execution.

02 Agent Used by the Agent —— Agent Autonomy

This, we believe, is the true dividing line between “toys” and “productivity.” When a task becomes so complex that it requires invoking multiple sub‑capabilities, spanning several systems, and unfolding across multiple steps, a single agent simply cannot handle it. At that point, what’s needed is multi‑agent orchestration: a central agent decomposes the task, delegates it to several subordinate agents, each of which assumes a specific role, with the agents communicating via structured protocols. 
 

The warehouse Agent invokes the inventory‑counting Agent, the storage‑capacity‑analysis Agent, and the relocation‑recommendation Agent, then aggregates the results for the operations team—this is a case of an Agent invoking other Agents. 
 

It addresses the issue of “autonomy.” Humans no longer need to instruct machines step by step; they simply need to specify the goal and the constraints.

03 Human-Used Agent —— Human-AI Collaboration

This is a category that has long been largely overlooked, yet whose commercial value remains severely undervalued. For the vast majority of enterprises, WMS, TMS, and ERP systems cannot be simply torn down and rebuilt overnight—they underpin core business operations, and the cost of replacement is prohibitively high.

But that doesn’t mean AI can’t get in. Through the MCP standardized connectivity protocol, we “agentize” existing systems: without altering their original architecture, we enable agents to interact with these systems just like new employees—checking inventory, creating shipping orders, updating statuses, and issuing commands. For the customer, the WMS they’ve been using remains the same; yet starting today, that WMS has a “thinking add-on.”

It addresses the “stock” issue, determining whether AI can truly be integrated into the day-to-day workflows of most enterprises.

04 Agent: The final category, natively designed from the ground up for agents.

It’s not “legacy system + AI”; rather, it treats the Agent as a first-class citizen in its design: the object model, access control framework, audit logs, and exception handling are all rearchitected around the premise that Agents will run here.

Within Deep Insights, the DI.AI Platform serves as precisely such a foundational layer—it is not merely an upgraded version of traditional SaaS, but rather an operating system designed specifically for a team of AI‑powered digital employees to “work” with.

It addresses the “future” challenge. In five years, newly launched supply-chain software will very likely be natively built on agents.

These four categories of challenges are not mutually exclusive. A mature enterprise AI strategy must successfully navigate all four pathways: leveraging human‑machine collaboration to secure quick wins, deploying agents to tackle complex scenarios independently, using agents to enhance and revitalize existing systems, and architecting agent‑native solutions to lay the groundwork for the long term.

Our Understanding of Agents (Part 2): Real-World Implementation—That’s the True Moat. Having discussed the technical roadmap, we must now turn to practical use cases.

Because we’re acutely aware of one thing: a general-purpose large model, even paired with a few carefully crafted prompts, simply can’t break through the real‑world complexities of supply chain operations.

Why “46 Agents ≠ a Large Model in Disguise” 
There are many solutions on the market that essentially just slap a generic large model into a conversational interface and call it an “industry‑specific agent.” While these offerings may look impressive during the proof‑of‑concept phase, they quickly reveal several fundamental issues once deployed in real‑world customer workflows: 
· It lacks an understanding of the business logic behind WMS, TMS, freight forwarding, and shipping, and cannot even clearly explain “how many parties and actions are involved behind a single bill of lading.” 
· It cannot connect to the customer’s corporate intranet—unable to view storage locations in the WMS or retrieve waybills in the TMS; 
· It cannot ensure SLA compliance or business regulatory adherence, and in the event of issues, there is no way to trace the root cause or provide accountability. 
· The bottom line is: there are plenty of pilot programs, but very few actually get implemented. 
 

The path to洞隐 is entirely different. Our 46 agents were “grown” over 22 years of industry data and business processes, through four key factors: 
Internalization of Industry Know-how

Over 55 industry-specific modules and more than 90 scenario‑specific knowledge graphs have been integrated into Ontology’s unified object model, enabling AI to truly understand the structured relationships among orders, goods, vessel schedules, and cargo space.

MCP Standardized Connection

WMS, TMS, ERP, customs APIs, carrier APIs, and IoT gateways are not customized for every customer; instead, they are integrated through standardized, protocol‑based interfaces.

Bounded Autonomy (Autonomy with Boundaries)

AI operates within clearly defined business boundaries, automatically escalating to human review when those boundaries are crossed. All actions are auditable and traceable—there is no black box.

Scenario-based specialized Agent

It is not a monolithic, all‑in‑one agent; rather, it comprises specialized agents—such as warehouse management, transportation, freight forwarding, shipping, customs clearance, and business operations—each performing its own distinct function and collaborating in a manner akin to real‑world digital employees. The entire platform is underpinned by a five‑layer architecture that provides tiered support from the foundational layer to specific use cases, seamlessly integrating the platform, connectivity, capabilities, orchestration, and end‑use scenarios into a closed-loop system.

A realistic slice of the scene in action

Let me mention a few agents who are currently on-site at the customer’s location.

1. Storage Capacity Analysis Agent

Real-time access to WMS location data enables automatic identification of high‑turnover and low‑turnover SKUs, with intelligent recommendations for optimal shelf locations. This boosts warehouse space utilization by 20%–35%, reduces the risk of stockouts by 40%, and shifts inventory‑capacity alerts from T+1 to real time.

2. Inbound Planning Agent

Automatically analyzes the scheduled arrival time window, allocates dock spaces and unloading resources, and provides real-time alerts for anomalies. Inbound waiting times are reduced by 50%, dock utilization increases by 60%, and scheduling is automated 24/7.

3. ABC Diagnostic Agent

Dynamically update ABC classification, identify risks of slow-moving, expired, and overstocked inventory, and automatically generate inventory optimization recommendation reports. Turnover rate improves by 30%, and inventory carrying costs decrease by 25%.

4. Intelligent Load Planning Agent

Replacing manual tasks such as vehicle scheduling, route selection, and time‑slot monitoring, it enables end-to-end order creation with a single click. Picking efficiency improves by 30%–50%, and new‑employee training is shortened from two weeks to just two days.

Behind these figures lies a clear value curve:

• ·Over 50% reduction in repetitive manual labor—robots never take time off, never get tired, and never make mistakes;
• Over 95% improvement in process efficiency—reducing turnaround time from hours to minutes and shifting from passive to real-time.
• 99%+ SLA compliance—AI‑driven automated assurance ensures no critical metrics are ever missed.
• Over 80% reduction in abnormal losses—shifting from reactive firefighting to proactive prevention.

The future of software is “interfaceless”—taking the rate‑negotiation agent as an example.

Regarding agents, the key point I’d like to emphasize at the end is this: the true way AI agents are transforming supply chains isn’t merely by making existing software smarter—it’s by gradually rendering the very notion of “software” obsolete.

The interaction logic of traditional software follows this pattern: “the user opens the system, clicks buttons, fills out forms, and submits workflows.” Over the past three decades, the evolution of enterprise software has essentially focused on optimizing this process—delivering better UIs, reducing the number of clicks, and streamlining workflows. Yet it has never moved beyond the fundamental framework of “people using software.”

The true paradigm shift brought about by Agents is that users no longer need to launch applications.

Our recent rate‑negotiation agent, which we deployed at a customer site, served as a validation of this approach.

In the past, a tender for full-truckload transportation followed roughly the following procedure:

• The operations staff opens TMS.
• Export historical freight rates and open email
• Contact twenty carriers to consolidate quotes in Excel.
• Open OA to initiate the approval process, then return to TMS to maintain the new contract.

Each round of communication, every iteration of price adjustments, and every step of approval tracking involves extensive system switching and manual interventions. For a medium-sized tender, on average, it consumes the primary effort of 2–4 full-time staff members, with a typical duration exceeding two weeks.

After integrating the access‑rate negotiation agent, the process has been streamlined into a single instruction: “Initiate the annual tender for Line XX, with a cost‑reduction target of 5%, deadline by next Friday.”

From this point forward, the Agent handles everything independently: it seamlessly integrates data across TMS, the contract system, the email system, and the OA platform; automatically conducts multi‑round negotiations with carriers according to our preconfigured negotiation strategies; feeds the negotiation outcomes back into the contract system’s approval workflow; and, in accordance with predefined rules, escalates any exceptions to human review. Throughout the entire process, the operations team never opens a single software interface—its only interaction is with the Agent’s chat window.

The results are as follows: reconciliation efficiency has improved by more than 95%, anomaly-related losses have been reduced by 80%, operations run unattended 24/7, and monthly labor costs have been cut by 2–4 full-time staff. Yet what matters even more than these figures is that they illustrate the future shape of supply-chain software: 
· From “interface” to “dialogue”—users no longer interact with the system by clicking, but rather by articulating their intentions to drive it; 
· From “process” to “goal”—people no longer need to break down every step; they simply need to articulate the objective and its boundaries. 
· From “systems” to “employees”—every component that was once a standalone software application will gradually evolve into a digital employee that can be “hired.” 
 

This is precisely why we define our product as an “AI Digital Employee,” rather than an “AI Feature” or an “AI Module”—because we believe that in the future, companies will employ not just humans, but also a team of AI employees who are always online, never take time off, never tire, and can be rigorously evaluated.

Written at the end

We don’t just slap a generic large model on and sell it to customers—we’ve trained an AI that truly understands supply chains, leveraging 22 years of industry data.

Deep Insights leverages the DI.AI Platform as its intelligent foundation and AI Agents as its execution units, covering four key scenarios—warehousing, transportation, freight forwarding, and shipping—to equip every link in the supply chain with autonomous closed-loop capabilities for perception, decision-making, action, and self‑healing.

If you’re also wondering how AI agents can truly integrate into your supply chain operations, you’re welcome. Click here to schedule a product demo · Schedule a free AI Agent consultation for shipping rates or warehouse diagnostics. 
Make the AI Agent a core member of your supply chain team.

@Deep Insights is an AI‑agent‑driven intelligent supply chain technology service provider that has been deeply engaged in the logistics sector for over 20 years. Leveraging the DI.AI Platform as its intelligent foundation and AI Agents as its execution units, it covers four major scenarios—warehousing, transportation, freight forwarding, and shipping—empowering every link in the supply chain with autonomous closed-loop capabilities for perception, decision‑making, action, and self‑healing.