The Next Software Paradigm in Automotive: The Rise of AI Agents

Software as we know it is undergoing a seismic transformation.

A few months ago, Microsoft CEO Satya Nadella sent shockwaves through the tech industry with a bold prediction for the coming era of AI Agents: "SaaS is dead."

“I think the notion that business applications exist that's probably where they'll all collapse right in the agent era let that sink in for a moment the CEO of the company that brought us Windows Office and basically created the software industry as we know it. ”

Ever since 1969, when IBM first unbundled its software and services from its hardware sales, the form of software has been in constant evolution.

In the traditional software era, companies sold perpetual licenses. Customers paid a large, one-time fee to have the software deployed on their own local servers and could, in theory, use it forever. But the true cost was far higher, as it came with the added burdens of expensive training, maintenance, technical support, and manual upgrades.

Then, in the early 2000s, a convergence of advancing computer power, widespread internet access, and new market demands gave rise to Software-as-a-Service (SaaS). Pioneered by companies like Salesforce (founded in 1999 on this very principle), the SaaS model changed everything. Standardized software was developed, maintained, and delivered to customers over the internet on a pay-as-you-go basis, dramatically lowering the cost and complexity of adoption.

For the last two decades, SaaS has dominated the enterprise landscape. Its low cost, easy deployment, and scalability made it the default delivery model, permeating every corner of business operations. From sales and marketing to HR, finance, and IT, there is a SaaS product for nearly every need.

But this traditional SaaS model has always had inherent flaws: rigid functionality, siloed data, and inflexible, pre-programmed responses.

Fast forward to 2025—what many are calling the dawn of the AI Agent era. Fueled by the explosive growth of Large Language Models (LLMs), a new form of software is emerging, promising even greater efficiency at a lower cost.

In Nadella's view, most SaaS applications are just elegantly packaged interfaces for database operations—a CRUD ("Create, Read, Update, Delete") front-end layered with business logic. He argues that in the near future, all this logic will migrate into AI Agents. These Agents will work seamlessly across multiple databases and systems, not just to complete simple tasks, but to handle complex business logic, and even write their own code to achieve a goal.

This vision is why companies are rapidly exploring new, Agent-driven ways of working. The capability boundaries of Agents are expanding daily, from perception and decision support all the way to autonomous task execution. Underscoring this trend, Gartner named Agentic AI one of its "Top 10 Strategic Technology Trends for 2025," predicting that by 2028, a third of all enterprise software will use Agentic AI, and 15% of daily work decisions will be made autonomously by Agents.

Just as the internet reshaped countless industries, the era of Agentic AI presents an opportunity for every business to "rebuild everything" with AI, unlocking unprecedented levels of efficiency and value.

How AI Agents Are Driving Fundamental Changes Across the Automotive Industry

The automotive industry is poised to be one of the most profoundly impacted sectors by this shift.

At its core, an automotive AI Agent is an intelligent entity that combines Perception, Reasoning, and Action, enhanced by Tools, Memory, and a continuous Cloud Feedback Loop. The fundamental concept is to leverage Large Language Models as a foundation, but augment them with planning capabilities, persistent memory, and the ability to orchestrate tools. The result is an agent that can actively perceive its environment, make intelligent decisions, and execute tasks autonomously.

The automotive world is a perfect crucible for this technology. It's an ecosystem defined by rich data streams, a huge diversity of complex scenarios, mature technology stacks, high market demand, and fierce competition. These are the ideal conditions for AI Agents to deliver transformative value.

Think of an AI Agent as a "digital employee" that can perceive, reason, and act. Its arrival on the scene doesn't just redraw the boundaries of what a vehicle's systems can do—it redefines the role of people. By taking over concrete operational tasks, Agents elevate human workers, freeing them to focus on strategic oversight, innovation, and high-value decision-making.

This change promises more than just significant efficiency gains in R&D, manufacturing, and service. It is set to catalyze the emergence of entirely new business models and organizational structures.

dentifying High-Value Application Scenarios for AI Agents in Automotive

Successfully implementing an AI Agent isn't about technology for technology's sake; it's about fundamentally re-examining business problems from the ground up. For automakers, simply "plugging in" an all-powerful Agent and hoping for the best is a recipe for failure. The right approach is to return to core business needs and target specific, high-impact scenarios where value can be created quickly.

So, what does a high-value scenario look like? Drawing from industry analysis, such as the "Enterprise AI Agent Value and Application Report" by JZ-Insight, the best opportunities share a few key characteristics:

• High Business Value: The potential ROI must be significant. Does the application directly improve a critical KPI or advance a core strategic goal?

• High Data Availability: The necessary data must be comprehensive, available in real-time, and clean. An Agent cannot function effectively on incomplete or poor-quality data.

• High Process Fit: The Agent must integrate seamlessly into existing workflows without creating high adoption costs or disrupting established processes.

This demands that automakers master two things. First, they must "Know Why"—developing a deep understanding of nuanced industry and user needs to identify the true nature of a problem and its potential value. Second, they must "Know How"—possessing the project management and resource orchestration skills to embed the Agent into the right process points and create a closed-loop system.

Through our partnerships with numerous automakers, we at EXCEEDDATA have identified many such opportunities, including intelligent diagnostics, smart thermal management, and AI-powered climate control. In these well-defined scenarios—where the problem is clear, the data foundation exists, and the process fit is high—the ROI of an AI Agent in terms of efficiency gains and cost reduction becomes immediately visible.

Let's take vehicle diagnostics as a prime example. In an era where brand reputation is highly sensitive to online sentiment, the speed and accuracy of fault diagnosis directly impact customer satisfaction. Yet, traditional diagnostics still rely heavily on manual experience and static rules, leading to slow response times, complex cross-departmental communication, and a tendency to miss early warning signs.

Now, imagine deploying a dedicated "Diagnostic Agent" capable of:

• Aggregating tens of thousands of real-time vehicle parameters.

• Integrating heterogeneous data, including fault codes, work orders, and historical repair logs.

• Performing multi-turn contextual analysis and reasoning.

• Automatically generating structured diagnostic recommendations.

• Continuously feeding its learnings back into the central knowledge base.

For an OEM, this would cause after-sales response efficiency to skyrocket while proactively managing brand risk.

The value of this scenario is clear, and many companies have already begun exploring it. This year, EXCEEDDATA launched its VDM (Vehicle Diagnosis Mgmt) system, which can identify 92% of pre-service faults and has helped clients reduce customer complaints by over 50%, delivering tangible, quantifiable returns on after-sales operations.

Looking ahead, VDM 2.0 will more deeply integrate AI capabilities to create a seamless flow: Natural language interaction (from a customer complaint or a technician's query) → Multi-turn dialogue combined with diagnostic model calls → Problem analysis and reasoning → Generation of a diagnostic suggestion → Continuous iteration of the diagnostic models and knowledge base.

The conclusion is clear: in the right context, AI Agent implementation is not just feasible—its value is undeniable. The automotive industry, with its dense data streams and complex processes, represents a vast, untapped landscape of opportunities waiting to be discovered, validated, and scaled.

What Automakers Must Build to Deploy Vertical AI Agents？

Taking thermal management and diagnostics as examples, implementing an intelligent scenario via an AI Agent requires the following core elements:

I. Data Infrastructure

1. Multi-Source Data Collection Capability

• Sensor data (temperature, flow, pressure, current, voltage, etc.)

• In-vehicle bus data (e.g., CAN, LIN, FlexRay)

• Historical diagnostic data (DTCs, work orders)

• Environmental information (weather, location, traffic conditions)

• Occupant information (expressions, status, clothing, behavior, etc.)

2. Multi-Modal Time-Series Data Management Capability (Database)

• High-efficiency processing with a time-series database (e.g., in-vehicle TSDB) for millisecond-level sampled data, log data, and vectorized feature data.

• Capabilities for data compression, indexing, and query optimization.

II. AI & ML Model Capabilities

1. Model Training Capability (Cloud-Side)

• Utilize historical big data to train multi-task models, such as:

  • Thermal management prediction models (cooling demand, fan control, driving behavior & trip)

  • Fault prediction models (battery, powertrain, thermal system)

  • Anomaly detection models (e.g., AutoEncoder or time-series anomaly detection)

2. Edge-Side Inference Models (Edge AI)

• Model quantization and pruning (e.g., TinyML, TensorFlow Lite).

• Run lightweight models on an SoC or MCU for real-time response.

III. Agent Logic and Framework

1. Basic Agent Capabilities:

• Perception: Continuously monitor vehicle status.

• Reasoning: Based on model outputs, a rules engine, and empirical reasoning.

• Action: Execute control commands, such as adjusting fans or issuing maintenance alerts.

• Learning/Adaptation: Continuously iterate models in the cloud to improve the Agent's intelligence.

IV. Fusion of Edge Computing Engine and Expert Knowledge

• Certain critical diagnostic logic must still be combined with expert know-how.

  • For example, a specific fault might require 5 consecutive samples to trigger, or simultaneously require Temperature > 70°C, an abnormal current, and a duration of 10s.

• This can be achieved via a Hybrid Approach: AI Model + Symbolic Rules.

V. Vehicle-Cloud Integration Mechanism

1. Closed-Loop Mechanism (Second-level deployment of diagnostic models across vehicle-cloud + flexible data collection)

• Support for model development in the cloud with second-level deployment to the vehicle.

• Support for the full cycle: Agent inference on the edge → Analysis and confirmation in the cloud → Manual review / feedback for training → Model updates.

• Fault localization accuracy improves over time.

VI. Explainability and Safety/Compliance

• Provide an explainable basis for every fault diagnosis.

• Meet automotive software safety standards (e.g., ISO 26262, ASPICE).

• Ensure sensitive data does not leave the vehicle (via edge database + edge computing) to protect personal privacy.

Conclusion

In any vertical industry, AI Agents represent a revolutionary leap. They fuse data, domain expertise, business logic, and large models into intelligent systems that drive decision-making and enhance the end-user experience. This is the defining trend of the Agentic AI era.

However, there is a major barrier to entry. Truly effective AI Agents must be built upon a deep "digital foundation" of high-value datasets and specialized domain knowledge—a foundation that can take years to build. The reality is, very few automakers are truly ready to capitalize on this shift on their own.

This is precisely the gap EXCEEDDATA was built to fill. We build the AI Data Infra for intelligent vehicles, solving the foundational data challenges that prevent automakers from deploying powerful AI Agents. Our infrastructure is built around three core technologies: our on-vehicle multi-modal time-series database, vehicle-cloud federated computing, and advanced Edge AI.

Today, our AI Data Infra is already deployed or in active deployment in nearly one million vehicles across more than 30 production models from 10+ of China's leading automakers.

Building on this proven foundation, EXCEEDDATA and our partners are now creating a full-stack solution—from vehicle-cloud computing and the business knowledge engine all the way to the AI Agent itself. Our mission is to empower automakers to move beyond theory and fully embrace the power of the Agentic AI era.