Core model
Human → Intent → Agent Computer → Output
What are Agent Computers?
Agent Computers:
The PC Era, Amplified
Agent Computers is a new category of devices built to run your AI agents full-time. They can sit in your home or office, always on, always available, always working.
You do not operate them like a PC. You delegate to them. You send a message on your preferred messaging app. Your agent gets moving.
The Evolution of Computing
Agent Computers
Built for AI agents to run continuously
No monitor required — sits quietly in home or office
You delegate tasks; agents execute autonomously
Always on, always working — even while you sleep
Interface is wherever you are: WhatsApp, Slack, Message
Runs your agents
Personal Computers
Built for humans to operate directly
Screen, keyboard, mouse — always present
You open apps and issue commands one step at a time
Machine waits for your input to proceed
Interface is the box in front of you
Runs your apps
PCs remain the machines you use. Agent Computers become the machines that work for you.
What Will Agent Computers Do for You?
Imagine waking up to find that while you slept, your agent has already monitored your key business metrics, flagged the three things that need your attention today, drafted responses to your most time-sensitive messages, and assembled a briefing for your first meeting – complete with the latest data and relevant context.
Agent Computers do not replace what you can do. They multiply it.
Professionals
Extraordinary leverage — operate at a scale and pace that wasn't achievable before. Research, analyze, and synthesize while you focus on decisions that matter.
Creators
More time for original work, less time managing logistics. Let agents handle scheduling, sourcing, formatting, and distribution so you can stay in the creative zone.
Developers
A local AI environment with privacy and control at the center. Run sophisticated models, build and test agents, iterate fast — without cloud latency or per-token costs.
Small Businesses
Operational capacity that used to require a much larger team. Monitor, plan, report, and execute — with agents that work the night shift so you don't have to.
Always-On Research
Agents that monitor news, markets, competitors, and signals 24/7 — surfacing what matters, filtering what doesn't, and briefing you when you need it.
Personal Delegation
Send a message, assign work, ask for status — from WhatsApp, Message, Slack, or wherever you already are. The interface is you, not the box.
AMD: Built for the Agent Era
Systems powered by AMD Ryzen™ AI Max+ processors, including the AMD Ryzen™ AI Max+ 395, are exceptionally well suited to become Agent Computers. They have the horsepower to run sophisticated local models, the efficiency to stay persistently available, and the architecture to support the parallel, multi-agent workloads that define what comes next.
Performance vs. Competition
LLM Model Size Support (up to 200B params locally)
Read related blogs for Stable Diffusion, and Token Throughput.
128GB
Unified Memory – enables running up to 200B parameter models locally
256GB/s
Memory bandwidth for sustained AI workloads
16 cores
Zen 5 cores
50+ TOPS
Dedicated NPU for continuous AI inference
Agent Ready AMD Systems
How to Build your own Agent Computers
Read our technical guide on how to configure your own Agent Computers to run OpenClaw Locally On AMD Ryzen AI Max+ And Radeon GPUs.
Frequently Asked Questions
At AMD, we describe an Agent Computer as a system built to run AI agents continuously. Rather than serving as the screen-keyboard-and-mouse computer a person uses directly all day, its role is to provide dedicated compute to the AI Agent who works on your behalf.
No. A person may help set it up, but after that the system is meant to be used primarily by the agent or agents. A traditional PC is something you use directly, while an Agent Computer is designed to run agents that use apps and services for you.
Not necessarily. An Agent Computer can operate without the usual PC peripherals once setup is complete. Because its resources are dedicated to the AI agent, it does not need to behave like a desk-bound system that a person is constantly sitting in front of.
We see people interacting with the AI agent through the communication channels they already prefer, such as WhatsApp, Slack, or similar interfaces. You talk to the agent like you would talk to any other person and the agent uses the Agent Computer to do the work, and it comes back with results, updates, or follow-up questions.
At AMD, we see Agent Computers as well suited for ongoing, multi-step work rather than just one-off prompts. That can include researching information, coordinating tools, organizing actions, preparing summaries, monitoring workflows, and continuing work even when you are away from your desk.
From AMD's point of view, the difference is not simply that the system is small or always on. The difference is that it has the hardware to run AI agents. We think of an Agent Computer as being defined by that role, along with hardware characteristics that are important for AI inference, such as strong compute and large unified memory for agentic workloads.
Local AI is important because it can give users more control over privacy, more predictable costs, and an always-available foundation for agentic computing. We also make room for hybrid approaches, so an Agent Computer can work locally when control matters most and still tap cloud resources when more scale is needed.
We highlight AMD Ryzen AI Max+ processor-based systems because we believe they are a strong fit for Agent Computer workloads. In AMD's view, they bring together local AI performance, efficiency, and an architecture that is well suited for more demanding inference and multi-agent scenarios, including Windows-based systems in this category.
Today, systems powered by AMD Ryzen AI Max+ processors as strong examples of what can fit this role, with partners such as Acer, ASUS, Framework, and Corsair building systems that can support this model.
Footnotes
CAUTION/DISCLAIMER: OpenClaw is a highly autonomous AI agent. Giving any AI agent of this nature access to any system may result in the AI acting in unpredictable ways with unpredictable/unforeseen outcomes. Use of any AMD suggested implementations is made at your own risk. AMD makes no representations/warranties with your use of an AI agent as described herein. Failure to exercise appropriate caution may result in damages (foreseen and/or unforeseen). To help protect from any such damages, we recommend the use of the following precautions PRIOR to use of any AI agent of this nature:
- Run the AI agent on a separate, clean PC that has no personal data, or within a virtual machine. Only ever copy specific data you want the agent to access.
- Strictly avoid granting access to any of your personal or primary accounts. Instead, create dedicated accounts for the agent and provide only the specific information or permissions required for these separate accounts.
- Carefully review and limit which skills or extensions are enabled for the agent.
- Ensure that any interfaces used to access your agentic assistant (such as the web UI or messaging integrations) are protected and not accessible without authorization over local networks or the internet.
Footnotes
CAUTION/DISCLAIMER: OpenClaw is a highly autonomous AI agent. Giving any AI agent of this nature access to any system may result in the AI acting in unpredictable ways with unpredictable/unforeseen outcomes. Use of any AMD suggested implementations is made at your own risk. AMD makes no representations/warranties with your use of an AI agent as described herein. Failure to exercise appropriate caution may result in damages (foreseen and/or unforeseen). To help protect from any such damages, we recommend the use of the following precautions PRIOR to use of any AI agent of this nature:
- Run the AI agent on a separate, clean PC that has no personal data, or within a virtual machine. Only ever copy specific data you want the agent to access.
- Strictly avoid granting access to any of your personal or primary accounts. Instead, create dedicated accounts for the agent and provide only the specific information or permissions required for these separate accounts.
- Carefully review and limit which skills or extensions are enabled for the agent.
- Ensure that any interfaces used to access your agentic assistant (such as the web UI or messaging integrations) are protected and not accessible without authorization over local networks or the internet.