Менеджеры автопарков внедряют «вайб-кодинг» для автоматизации учета и оптимизации ресурсов

Of course, this takes time, money, and resources that demand a well-defined ROI. That’s all changed. Ernie Garcia, director of fleet and business systems for Gothic Landscape, is building a spec’ing program that combines information from several spreadsheets and automakers’ order guides.   “It allows us to build specs and quickly update them, kind of like little Tetris pieces,” said Garcia. He can add pricing, customize the output, and produce a version fleet managers can print, while hiding pricing when necessary. Garcia didn’t learn Python, JavaScript, or another programming language. Using vibe coding, he described the problem in plain English and worked with an AI model to create the solution. Vibe coding — you’ve no doubt heard that phrase by now — is spectacularly disrupting traditional processes by using AI to build a solution in-house, at a desktop, with a negligible cost, and in a fraction of the time. Vibe coding allows someone without a traditional software-development background to tell an AI system what he or she wants to accomplish. The AI generates the underlying code, spreadsheet, application, or workflow. The user tests it, explains what needs to change, and continues refining it. This isn’t theory. As Garcia can attest, it’s happening right now. Can We Finally Drink From the Hose? The phrase, “Drinking from a firehose of data” isn’t new. For years, this firehose has been blasting from telematics platforms, fleet management information systems, OEM portals, CRM systems, maintenance records, fuel transactions, vehicle valuations, and other internal systems. As the hose of information amplifies, fleet managers are expected to accomplish more with fewer people. Garcia manages his fleet in-house, without a fleet management company. Vibe coding helps bring it all together at scale with the same personnel. Tackle Small Projects First However, our operators recommend not to start vibe coding by drinking from the hose. In fact, it’s best to start with a manageable, manual application. Garcia vibe-coded a small application to batch-print dozens of PDF files. He’s expanded into bigger problems, using multiple large language models (LLMs) to help. Garcia needed a total cost of ownership calculator that his fleet managers could easily use. He described his users as "mechanics first and not really analysts." He used ChatGPT to help develop the prompt, then provided that prompt to Copilot, which produced a spreadsheet with multiple tabs and TCO calculations. In other words, one AI helped him write better instructions for another AI. Garcia also uses AI for policies, procedures, SOP reviews, and comparing agreements. He estimated that one type of task that previously took as much as 80 hours might now take him roughly 30 minutes. The Fleet Manager Still Makes the Decision Reed Jackson, fleet analyst for Dallas County, is exploring AI around vehicle replacement and electrification. His fleet replacement analysis requires pulling together vehicle age, mileage, maintenance history, utilization, and other factors. Jackson is working on using those inputs to generate a priority-ranking system that can help identify which vehicles warrant attention. "The LLM isn't going to make a decision for you," said James McKinley vice president of operations at City Rent a Truck, but it can make the decision-making process considerably faster by consolidating information that otherwise might come from five or six sources. McKinley's operation is examining ways to combine total cost of ownership, replacement cycles, and residual values. In the past, gathering that information across thousands of assets could require analysts or data engineers. AI tools can dramatically reduce the barrier to bringing those inputs together. McKinley knows the problem, but doesn’t have a software engineering background. Therefore, he doesn’t have to explain the problem to a programmer and hope the programmer translates correctly. From Spreadsheets to Fleet Applications Kristin Leary, director of fleet services at Syneos Health, offers another example of the kind of fleet problem suited to this approach. Syneos Health handles temporary projects and rotating employees, which creates an ongoing vehicle-assignment challenge. When a new territory needs a vehicle, the fleet team must determine which available unit makes sense based on factors such as location, vehicle tier, drivetrain, and how long the vehicle has been sitting in storage. Today, much of that process involves spreadsheets. Leary's team is developing an application in which a user enters the location and type of vehicle needed. The system would then identify the closest appropriate vehicle while prioritizing units that have been sitting in storage longest. A second app in development would address another familiar problem: vehicles being returned without keys, plates, or current registration, or with damage that wasn't previously reported. The app would periodically guide drivers through photographing the vehicle's exterior, interior, license plates, and other items, creating visibility into vehicle condition before the unit eventually comes back for sale. Leary sees AI's value in automating the tasks that prevent employees from spending time on higher-level work. "If we can automate some of the tasks that we do on a day-to-day basis and enable our team members to do a lot more of the strategic work that can't be handed off," the organization can concentrate on the areas where its people have specialized knowledge, she said. The New Skill — Prompt Engineering Building with AI requires a new kind of skill — prompt engineering — that is different from traditional programming. "What you put in is what you're going to get out," McKinley said. One of his techniques is to ask the AI itself to act as a prompt engineer and revise his instructions for the best possible result. Then he continues the conversation, refining what he wants until the model has a clear set of instructions. The workflow may look more like this: Describe the problem. Review the result. Explain what is wrong. Add a requirement. Change a calculation. Test again. Refine the output. Save the successful process and reuse it. Don Schnell, a business consultant and former member of the Verizon fleet, compared that process to writing and revising a document. Once the instructions and output are correct, the routine can be saved and run again rather than rebuilt each time. Start Small, With Something You Understand For fleet managers who haven't built anything with AI, the most useful first project should be intentionally easy to grasp. Garcia suggests starting with basic fleet data. Using an approved AI environment, take information you already know well and ask it to calculate or organize familiar metrics: fleet age, maintenance, costs, and fuel economy. Starting small keeps the project manageable, but it also allows the user to already know what the results should look like. Fleet Knowledge Still Matters An important lesson in using this new generation of AI tools is that the people experimenting with vibe coding today learned their jobs before generative AI existed. They spent years working in spreadsheets, manually reviewing maintenance data, building replacement analyses, and learning through experience when something didn't make sense. But future employees, Schnell warned, may encounter the AI interface first. If those users haven't learned to examine the data themselves, they may accept an AI-generated answer because the system presents it confidently. "We need those users to learn how to manually look at the data and understand what good data, what's bad data before they just solely rely on the results of AI," Schnell said. Failing to develop that foundation "would be a fatal mistake." That raises a vital question: What happens when the next generation learns the AI tool before learning the fleet? The answer shouldn't be to avoid AI. It should be to ensure that AI proficiency and fleet proficiency develop together.