AI Agents: What Are They, and How Will They Impact the AI PC Space in 2025?

What are AI agents?

Over the past several quarters, OEMs have focused on incorporating local AI-powered features into their new PC releases, with initial neural processing unit (NPU)-enabled use cases leveraging AI to further enhance collaboration experiences and extend battery life. However, AI agents take the NPU’s functionality a step further, combining the capabilities of large language models (LLMs) with other resources to partially or fully automate a wide range of tasks, including responding to emails, booking hotel stays, or opening and closing IT help desk tickets.
 
LLM-based AI systems have traditionally been programmatic in nature, making them well suited for accomplishing a relatively narrow range of tasks quickly but with a varying degree of accuracy depending on the user’s specific query and how closely it aligns with how the LLM was trained. However, as AI has matured, an increasing number of organizations have invested in the development of compound AI systems, making way for the rise of agentic AI. Compound AI systems, which include AI agents, combine AI models with additional resources such as adjacent AI models, external data sets, web searching capabilities and other APIs to address some of the limitations of programmatic AI systems. This allows AI agents to carry out more complex tasks with a higher degree of accuracy compared to traditional programmatic AI systems, like ChatGPT.
 

 
As a general rule, as AI systems become more compound, speed is sacrificed. However, the compound nature of AI agents is what allows them to act on behalf of the user — a key differentiator and the primary value proposition behind agentic AI. Without any human intervention, an AI agent can create several subtasks where it brings in and analyzes data from several sources before determining the next step in the process.
 
It is worth noting that while AI PC agents typically leverage the NPU, most AI PC agents do not operate completely locally, leveraging cloud computing resources.

Proprietary AI agents will become increasingly prevalent in the AI PC space over the next several quarters

Maximizing AI PC appeal through software integration

For OEMs to attract customers to their AI PC offerings, the devices must have software that leverages the power of the NPU in a way that improves performance, productivity and/or security.
 
One of the most important software solutions underpinning the AI PC space is Microsoft Copilot+, which offers a series of generative AI (GenAI) features and experiences for Windows 11 machines leveraging several of Microsoft’s small language models. However, not all Copilot+ functions are run natively on the device, with certain queries going to the cloud, which may lead to security and privacy concerns for some users.
 
To differentiate its AI PC portfolio and bring more AI tasks onto the device itself, Lenovo has developed an agent known as AI Now, which has capabilities such as document management, meeting summarization, device control and content generation. Leveraging a local large language model built with significant collaboration from Meta, AI Now offers enhanced data privacy and enables GenAI features without internet connectivity by allowing users to interact in real time with the device’s personal knowledge base, rather than relying on cloud computing.
 
With the release of its first set of Next-Gen AI PCs in May, HP Inc. announced a similar application to Lenovo’s AI Now, named HP AI Companion. Available for download on any HP AI PC with a 40-60 TOPS (trillion operations per second) NPU, the application leverages OpenAI’s GPT-4 model to bring AI tasks such as performance optimization, document summarization and content generation onto the device.
 

Click the image below to register for our upcoming TBR Insights Live session, AI PCs in 2025: Unlocking Mass Appeal and Overcoming Market Challenges

AI agents as differentiators in the AI PC market

We expect to see these offerings become increasingly central to the AI PC space over the next several months, with vendors tapping into buyers’ concerns about data privacy related to cloud computing in order to promote their own proprietary AI agents. Vendors will continue to position these agents as complements to Microsoft’s Copilot+, rather than replacements, as they will shy away both from attempting to compete with Copilot+ and other cloud-based offerings and from alienating Microsoft, a vital partner when it comes to AI.
 
Overall, these agents are currently being more heavily marketed toward commercial customers, as that subsegment of the market is generally more strategically valuable to OEMs because of commercial PCs’ higher average revenue per unit (ARPU) and attach rates for peripherals and services.
 
However, TBR expects these agents to gain traction in the consumer AI PC space as well, especially as they include features useful to all users, such as performance optimization and increased security, as well as those designed specifically for enhancing workplace productivity. Ultimately, TBR believes the extent to which vendors can educate users on when and how to use specific AI tools will determine the level of adoption of individual AI tools. The availability of multiple tools on the PC is likely to lead to confusion.
 
Similar to AI PC OEMs, smartphone vendors are becoming increasingly invested in baking AI into their devices to enhance the devices’ value and accelerate the refresh cycle. Many of these features revolve around photo and video editing software, such as Google’s Magic Editor and Photo Unblur, as well as notification and document summarization. Personal agents that allow the user to navigate their device through voice commands and natural-language text such as Apple’s Siri are also popular.
 
Smartphone vendors are also combining on-device and cloud-based AI processing when building out the functionality of these devices, with the recently released Apple Intelligence platform being the most prominent example. When possible, queries are processed on the device through the NPU built into Apple’s proprietary chips, while queries that are more complex are sent to the cloud through the company’s Private Cloud Compute system. TBR expects that this hybrid model will increase in popularity as devices vendors balance greater AI functionality with data privacy and security.

 

Learn more

Download 2025 Predictions special report: AI PCs: Progress, Potential and Hurdles in Redefining the Market in 2025
 
Attend TBR Insights Live session: AI PCs in 2025: Unlocking Mass Appeal and Overcoming Market Challenges