A single vehicle model contains hundreds of low-voltage wire harnesses. Parameters such as wire diameter, pin configuration, current load, voltage load, and temperature must each be calculated individually, a task that previously consumed two full days, or 2,880 minutes, for a seasoned engineer. Now, after importing a parameter table, the results are generated in roughly five minutes.

This tool was developed by Mao Hai, an electrical systems integration test engineer at Chongqing Changan Automobile Company Limited (000625.SZ). He converted wire harness selection rules into a product requirements document and then leveraged AI to complete the development. The AI handles the computation, analysis, and presentation, while engineers verify the process and exercise final judgment.

At Hebei Changan, quality engineer Guo Ziya has been tackling problems with the same methodology. A batch of barcode scanners could not directly input Chinese characters, forcing employees to manually type quality issues. Inconsistent phrasing across different staff members subsequently skewed data analysis and issue classification. Replacing the equipment would require additional capital, and developing a new system meant waiting for the IT development queue.

Guo decided to build a solution herself. She joined Hebei Changan in 2013 with a materials science background and no software development experience. Using AI, she first created a text conversion and QR code generation tool, then added an automated verification feature to accommodate the scanner's byte constraints. The system assesses whether a QR code can be recognized, highlights necessary adjustments, and finally generates a code compatible with existing hardware.

Building on that success, she extended AI usage to vehicle document inspections, automated submissions, and verification of root causes and corrective actions. Although unfamiliar with intelligent agent workflows, she methodically explained each actual process step to the AI, allowing it to map out how parameters should connect. A quality engineer with 13 years of experience has now begun directly modifying the workflows she uses every day.

The shift from 2,880 minutes to five illustrates how Changan is pushing AI to the operational frontline: the people who understand problems best are now the ones creating their own solutions. In recent years, external discussions about AI in the automotive sector have focused overwhelmingly on intelligent driving assistance and smart cockpits. However, Changan is rolling out Qianwen Office across its internal operations, embedding it into daily work spanning R&D, production, procurement, operations, and after-sales service.

The Qianwen Office deployment at Changan offers a compelling glimpse into a different dimension of enterprise AI adoption. Model capability is merely the starting point; who raises the questions, how clearly business rules can be articulated, and how individually built tools become integrated into the organization ultimately determine the real-world impact.

AI Thrives First in Well-Defined Workflows

Mao Hai chose low-voltage wire harness selection because it boasts unmistakably clear inputs, outputs, and calculation logic. Engineers must process parameters including wire diameter, pin configuration, current load, voltage load, and temperature, with a substantial computational burden, yet every step follows established rules. Mao worked alongside harness engineers to map out the calculation logic, documenting inputs, outputs, and rules in a product requirements document before tasking AI with developing an executable tool. After deployment, the tool was enhanced with a result traceability feature that lists every calculation step for engineers to verify.

The division of labor is straightforward: AI performs the calculations, while engineers retain responsibility for judgment and confirmation. Mao summarizes their roles directly: "Humans take charge of judgment and direction; AI handles computation, analysis, and presentation." This represents one of the easier scenarios for Changan to operationalize: recurring work, describable rules, structured data, and verifiable outcomes.

Xiao Shiqiang, Manager of AI Application Development at Changan, notes: "Scenarios don't require deliberate hunting. Business pain points have always existed; technology is simply the means to resolve them." Guo Ziya's vehicle document consistency checks also fit these criteria. Each vehicle leaving the factory requires over 100 mandatory inspection parameters on the certificate and accompanying documents to match national regulatory website records. Previously, staff had to manually review certificates, documents, and web data one by one. Now, after a photo is uploaded, the tool extracts parameters, cross-references announcement data, and flags discrepancies, allowing staff to focus mainly on exceptions. Data from earlier interviews indicates that per-vehicle review time has dropped from over ten minutes to just one or two.

The clearer the rules, the more reliably AI can execute. If processes still depend on individual experience, or if different departments interpret the same field differently, models cannot compensate for those organizational gaps on their own. Mao encountered this exact challenge during user VOC analysis. His team had to choose between two product proposals, with user feedback scattered across automotive forums, social media platforms, and comment sections. Engineers first defined keywords, classification standards, and output formats; AI then handled data collection, sentiment recognition, opinion clustering, and tabular organization. Work that previously required one to two weeks can now be completed in approximately one hour after process debugging concludes.

While AI has lowered the barrier to data processing and programming, business personnel must still answer two fundamental questions: which data can be trusted, and which results are suitable for product decisions.

Frontline Staff Build Their Own Tools, Shortening the Traditional IT Pipeline

Traditional digital transformation typically follows a lengthy handoff chain: business departments submit requirements, product managers interpret and translate them, technical teams or external vendors develop solutions, and after testing and acceptance, employees finally receive the finished product. With each additional handoff, more business nuance is lost.

Quality management serves as a textbook example. Guo Ziya deals daily with issue entry, data submission, root cause analysis, and corrective action tracking, and she knows exactly which stages are prone to fatigue, omissions, and data discrepancies. With AI available, she could quickly prototype a small tool and later integrate effective functions into existing platforms. System improvements that once required waiting are now being shaped directly by the people doing the work.

This shift does not eliminate the IT department's role. Business users can build small utilities, automation scripts, and product prototypes, but once tools move into production systems, professional teams are still required for interfaces, permissions, testing, security, and ongoing maintenance. Business staff have gained development capability, while IT teams increasingly focus on platform and governance responsibilities.

Gong Xuan, Senior Digital Project Manager at Changan's Global Procurement Platform, has pushed this transformation even further. Before 2025, she had no IT background and primarily worked in cost-related functions. By 2026, she had adopted an "AI First" mindset, automatically assessing which portions of any given task AI could handle before starting work. An AI-generated usage review revealed she has built 58 Skills, with over 240 interactions within two weeks, saving an estimated 45 hours according to the tool's calculations. While this figure comes from tool estimation and doesn't meet financial reporting standards, it clearly reflects AI's deepening integration into daily work.

Gong continuously organizes emails, meeting minutes, instant messages, and online documents into structured knowledge modules. AI uses this information to update project progress and responsibility assignments, generate daily reports, to-do lists, and meeting materials, and assist in diagnosing system issues and suggesting troubleshooting directions. This personal workflow system requires ongoing maintenance. Every week, Gong reviews existing Skills to determine which need adjustments or optimization. She describes her rhythm plainly: "Monday through Friday, I let AI handle the work; on weekends, I discuss with AI what areas still need improvement."

Previously, employees used software to complete tasks. Now, while software remains, AI is emerging as the gateway to accessing knowledge, files, and diverse capabilities. One of Qianwen Office's roles at Changan is enabling employees to invoke these capabilities within their existing work environment. Alibaba provides the foundational models, products, and platform support, while Changan retains ownership of its business processes, internal data, and application scenarios. For large enterprises, generic models can be purchased, but years of accumulated business context must still be organized in-house.

Frontline Identifies Problems, Transformation Teams Amplify Capabilities

An individual saving a few hours does not automatically translate into shorter departmental processes. Shu Junliang, Vice President of Qianwen Office and Head of Product and Research, offers a telling example. A process originally involved three people, each spending two hours. With AI intervention, each person might need only 15 minutes, yet the overall process still takes three hours because communication, waiting, and handoffs remain unchanged.

After individual efficiency gains, enterprises encounter more complex challenges: how to help AI understand each employee's department, projects, customers, and permissions; how to connect existing CRM, ERP, and internal platforms; and how to maintain workflow continuity between people and between humans and agents. At Changan, these horizontal responsibilities fall to teams such as the Transformation and Efficiency Department and the AI Application Development Division.

Business units remain closest to the problems. Horizontal teams handle tool introduction, development support, training and promotion, operational data, system integration, and security governance. Without both sides, applications struggle to scale. Leaving everything to technical departments risks reverting to the old requirement-queue bottleneck; granting employees unrestricted freedom creates duplicate builds, permission chaos, and abandoned maintenance.

Changan has not resorted to administrative mandates to enforce usage targets. According to Xiao Shiqiang, the company first encourages senior leadership to install and use the tools. Through personal demonstration, adoption gradually spreads organization-wide, helping everyone recognize that AI can genuinely solve their problems. Backend data also informs operational efforts. Teams seek out high-frequency users and invite them to share real-world case studies weekly. When usage appears low in a particular department, they engage directly to understand the reasons and provide training. Metrics such as daily active users, active days, Token consumption, and Skill counts are monitored, but these indicators are primarily used to assess whether employees are genuinely starting to use the tools.

Individual Skills are also being converted into organizational assets. Gong shares her professional Skills with colleagues. Since each person's computing environment and knowledge base differ, recipients must still tailor the tools to their own work. She estimates that others can quickly acquire roughly seventy percent of the capability from a Skill, with the remainder requiring personal refinement and judgment. In the past, experience primarily flowed from mentor to apprentice. Now, a portion of working methodologies can travel through Skills.

Replication still has its limits. A Skill that functions in the procurement center cannot necessarily transfer unchanged to manufacturing or sales departments. Field definitions, permissions, systems, and review responsibilities may all vary. The organization must preserve unified standards while allowing business units to continue adapting. Changan has thus developed a combined top-down and bottom-up approach: frontline employees start with small problems, and the corporate level then evaluates which applications deserve sharing, integration, and long-term maintenance.

As Token Bills Rise, Enterprises Must Calculate the Full AI Equation

Changan has not imposed rigid "cost savings or efficiency gains" targets on departments. Xiao Shiqiang's reasoning is straightforward. Enterprise AI remains in an exploratory phase, with rapid changes in application effectiveness and technological capability. Converting innovation prematurely into operational KPIs invites fabricated results. If product experience fails to meet expectations, forcing employees to boost usage frequency will only breed resistance.

At this stage, Changan prioritizes whether employees are actually using the tools and whether viable scenarios are emerging. Token and compute consumption are growing rapidly, a development the company anticipated. Management has already begun asking where these costs will be borne. As Xiao puts it: "Over the long term, AI must still demonstrate return on investment." That accounting will inevitably find its way into the cost structure.

When an employee uses AI to generate a presentation, the time saved is relatively easy to quantify. But once a tool becomes embedded in quality, R&D, or procurement workflows, the value calculus becomes far more complex. It may shorten delivery cycles, reduce errors, improve product quality, or make analyses feasible that were previously too costly to pursue. Simply counting invocation volumes cannot capture these shifts.

Data security presents another challenge. Changan manages data according to classification levels including internally controlled, standard commercial confidential, and core commercial confidential. Classified and core confidential information is primarily processed by internal AI agents, and the company is building private deployment capabilities. External services require agreements that define data boundaries; certain tasks run in local environments, and cached data must be cleared as specified after invoking cloud models.

Accuracy also demands accountability. After using AI to generate presentations, daily reports, and data analyses, Gong still reviews the results, feeds errors back to the AI, and adjusts the rules. Text-based tasks become increasingly stable through accumulated refinement; data-driven tasks, even when executed by code, still require verification of fields, calculations, and outliers. Changan now stands at the intersection of two phases. The first phase addressed whether anyone would use the tools, helping employees complete their first real problem. The second phase must answer whether continued investment is justified, translating daily active users, Token consumption, and Skill counts into process cycle times, quality improvements, and business outcomes.

For enterprises preparing to adopt AI, Changan's experience suggests a relatively clear sequence. First, identify work that is rule-bound, repetitive, and verifiable, then engage the people who know the problems best in development. Once scenarios prove viable, determine how to integrate systems, share capabilities, define permissions, and allocate costs. When frontline employees begin spontaneously building AI tools, management must catch those tools: select replicable scenarios, connect them to enterprise systems, clarify permissions and accountability, and validate investment through business results.

The journey from 2,880 minutes to five captures the efficiency of a single-person tool. How to transform one person's tool into organizational-wide capability is the equation Changan must now solve, offering a compelling benchmark for how very large enterprises can genuinely harness AI.