AMDWhile the discussion of AI has largely centered on its powerful models and capabilities, a new challenge is emerging: how to run AI agents efficiently at scale. As organizations embrace agentic AI, technology leaders are rethinking everything from infrastructure and computing strategies to the economics of enterprise AI. As enterprises move from AI assistants to autonomous AI agents, organizations must also rethink how enterprise AI workloads are deployed, secured and optimized across cloud and AI PCs.In conversation with Forbes, Rahul Tikoo, senior vice president and general manager of the client business unit at AMD, explains why AI agents are reshaping enterprise computing and how organizations are balancing cloud and local computing.The following interview has been edited for clarity and length.Why AI Agents Matter For Enterprise ComputingForbes: AI assistants have captured most of the attention over the past few years. Why do AI agents represent the next major computing shift?Tikoo: AI assistants are more of a reactive tool … AI agents are workers. You’re telling them what you want to do and they're working on it for you … In the chatbot or assistant era, maybe you get 20-30% more productive. In this era, you're going to get a major productivity multiplier. This isn't about replacing people with machines — it's about turning your existing team into a more efficient version of itself. Your people can suddenly accomplish work much faster and that's where you unlock new value for your end customers. It's going to create entirely new businesses — even single-person businesses that can scale very fast because they have agents working for them.Forbes: What’s the biggest misconception enterprises have about agentic AI?Tikoo: That agentic AI is just the next evolution of chatbots. It’s really not. Chatbots are interactive, but agents act. An effective agent will actually understand your context, reason, find information, invoke tools, work and continue to do that until they get to the outcome you’ve set up.Another misconception is that agentic equals model intelligence. Of course, intelligence is important — you need to have the right small language or large language models. The bigger challenge is creating the right secure sandbox for these agents to operate and then addressing how you govern these agents. How do you address privacy concerns in certain industries? How do you make sure that the economics work?Forbes: What are the most compelling productivity gains you’re seeing from AI-powered workflows, and how do agents expand these opportunities?Tikoo: We looked at AI through a persona-based lens, from office workers and road warriors to software developers, financial analysts and other power users. Across these groups, the gains are becoming very real.Based on our observations, in software development, for example, many enterprises have already seen 20%-30% productivity improvements, with AI helping teams generate code, accelerate development and enable early-career developers to perform more like experienced developers.Agentic workflows can expand those gains even further, potentially moving from 20%-30% improvement to a much larger productivity multiplier. That impact shows up in specific ways: workflows can sort information, route requests, manage approvals and coordinate across teams; research can be synthesized into insights and recommendations; operations can flag risks, bottlenecks and conflicts earlier; and decisions can be grounded in an organization’s own data, policies and workflows.Forbes: Where are you seeing the strongest early adoption among enterprise clients? Are any industries moving faster than others?Tikoo: Software development is one of the clearest early use cases. AI can generate code, accelerate development and help solve complex problems, making experienced developers more productive while enabling early-career developers to perform at a much higher level. As natural language becomes more of a programming interface, more employees can create software and automate work without needing deep expertise in specific coding languages.Customer service and business process automation are also moving quickly. AI assistants and agents can sort information, route requests, manage approvals, surface relevant data and coordinate follow-ups, turning routine assistant workflows into everyday enterprise tools. In sales, customer support and operations, that means faster decisions, fewer manual steps and more time spent on the work that drives outcomes.The same pattern applies in professional services. Doctors and lawyers want to focus on patient care and client service, but administrative work often gets in the way: billing, insurance and claims, documentation, follow-ups and back-office coordination. AI agents can prepare documents, reconcile claims, track actions and organize information so professionals spend less time on paperwork and more time applying their expertise where it creates the most value.The Economics of Agentic AIForbes: Agentic AI consumes dramatically more tokens than traditional chatbot interactions. How should enterprises think about the economics as they scale?Tikoo: All organizations are dealing with this right now and I can tell you that the cost of agentic AI is becoming a P&L line item in many corporations … You’re talking about hundreds to thousands of dollars per employee in using a particular agentic workflow.So that’s certainly a big discussion topic in boardrooms. But I think these organizations also realize that this is an important investment they must make to unlock enterprise value.They all want to build better products and new services — but increasingly this is also about staying competitive. If a competitor is adopting agentic AI quickly and capturing that 10x, or even 100x, productivity gain, those advantages compound fast, and it becomes very hard to catch up. So waiting is the real risk.The organizations that win will be the ones that move quickly and do it efficiently: routing each task to the right-sized model instead of sending everything to the most expensive frontier model, training their people on good AI habits and matching models to what each department actually needs rather than forcing one model to do everything.This is exactly where AMD comes in. As companies work through these challenges, we have the technology and the expertise to help them build the right hybrid infrastructure, put AI to work effectively and manage their costs as they scale.Forbes: One of the biggest debates is whether workloads should run in the cloud or on local devices. How do you see that balance evolving as agents become more sophisticated?Tikoo: Smaller, highly capable models are advancing rapidly. Models with just a few billion to tens of billions of parameters are now delivering performance that rivals much larger frontier models for many enterprise workloads.As these local models become more effective, agentic PCs will emerge as a strategic tier of enterprise AI infrastructure. They provide always-available, low-latency intelligence without the ongoing cost of cloud inference, enabling organizations to scale AI more broadly across their workforce.That doesn’t eliminate the need for the cloud. Frontier-scale models will continue to power the most complex reasoning, training and large-scale AI workloads. The future is not local or cloud — it's both.Leading enterprises are increasingly adopting a hybrid AI strategy, running everyday productivity, automation and knowledge workflows locally while leveraging the cloud for the most demanding tasks. This approach delivers the best combination of cost, performance, security and scalability.Meeting The Infrastructure DemandsForbes: Many people focus on graphics processing units (GPUs) when discussing AI infrastructure. Why does the central processing unit (CPU) remain a critical part of the agentic AI experience?Tikoo: The CPU, the GPU, the NPU [neural processing unit], the memory, the networking and software — all of it has to work together in an agentic AI workflow. The GPU is where you’re going to get performant AI. The NPU is where you're going to get efficient AI. But the CPU is what is going to help you from a control plane and orchestration perspective.Think of orchestration and the control plane as the coordination layer — the air-traffic control for the whole workflow. In an agentic task, the CPU decides which model or tool each step should use, routes the work to the GPU or NPU, keeps all the moving pieces in sync and pulls the results back together so the agent reaches the outcome you asked for. Without that coordinator, the rest of the hardware can’t operate as one system.Forbes: What role does memory play in enabling the next generation of AI agents, and what are the surrounding economics?Tikoo: As agentic workflows become more sophisticated, they will require larger context windows, multiple concurrent models and significantly more memory. This makes the AI PC a strategic part of enterprise AI infrastructure — not just another endpoint.For enterprises spending heavily on cloud AI, investing in agentic PCs can reduce inference costs while improving performance, responsiveness and privacy.The winning architecture will be hybrid: Agentic PCs handling everyday AI workloads locally, with frontier models in the cloud powering the most demanding tasks. Together, they deliver the optimal balance of cost, performance, at scale.Forbes: Looking three to five years ahead, what does success look like in the agentic AI era?Tikoo: It’s about unlocking the ability for people to do tasks they weren't able to do before. But it's bigger than just doing more. It's about accelerating outcomes — compressing what used to take weeks into days, and days into hours — and those faster outcomes compound into exponential results over time. You don't need to be experts in every area. The tools and the agents will be the experts for you, unlocking a new level of productivity so you can focus on what you want to accomplish and the outcomes you want.Forbes: What advice do you have for IT leaders who know this is where they need to be but aren’t sure where to start?Tikoo: Start by assessing where AI is already creating value in your organization. Identify the workflows that consume the most time, involve repetitive tasks or require employees to spend more effort on coordination than on high-value work. You cannot do everything at once, so prioritize a handful of use cases where AI and agentic workflows can deliver measurable productivity gains and business impact.At the same time, build the foundations for scale. Put the right governance, security, compliance and management frameworks in place so AI can be deployed responsibly across the organization. Treat AI as a business transformation initiative, not just a technology project.Most importantly, do not wait for the perfect strategy. Start with targeted pilots, measure outcomes and expand what works. The organizations that move now will build the skills, processes and competitive advantages that define the next generation of enterprise leaders. That is exactly the kind of challenge we help customers work through at AMD: assessing where they are, putting the right foundation in place and helping them move quickly and efficiently.Ready to put agentic AI to work? Explore how AMD Agentic PCs can help your business today.