Becoming a 10x Productivity Engineer

The most important shifts in computing rarely come from a single product or breakthrough. They happen when advances in technology, tools, and human behavior converge to fundamentally change what is possible.

We are at one of those moments now.

Agentic AI is moving beyond assisting with individual tasks. Increasingly capable systems can plan, write, test, debug and iterate across complex workflows, with people providing the goals, context and judgment. This is changing how software is developed, how engineering teams operate, and how organizations think about productivity.

I recently explored this transformation with Boris Cherny, creator and head of Claude Code at Anthropic, during an episode of AMD’s Advanced Insights. What struck me most was not simply how quickly coding agents are improving, but how deeply they are beginning to reshape the development process.

That is why the strategic partnership AMD and Anthropic announced in conjunction with Advancing AI is so significant. It connects leading AI models and developer tools with open, high-performance computing infrastructure, helping accelerate both AI development and the way we build the technology that underpins it.

From coding assistant to agentic collaborator

The first generation of AI coding tools largely focused on autocomplete: helping developers finish a line of code or generate a discrete function. That was valuable, but it did not fundamentally change the workflow.

The emerging agentic model is different.

As Boris explained, increasingly capable models can be given a goal, access to the right context and a set of tools and then determine how to execute the work. Instead of prescribing each step, engineers can focus on defining the problem, setting guardrails, and evaluating the result.

That represents another rise in the level of abstraction. Engineering has evolved repeatedly, from programming hardware directly, to punch cards, to higher-level languages and frameworks. Now we are moving from manipulating source code toward managing agents, workflows, and increasingly complex systems of agents.

At AMD, we are seeing this transition firsthand. Our inflection point came when coding agents began supporting complete agentic workflows rather than isolated portions of the development process. We are now applying these capabilities across software, firmware and aspects of chip design, including capturing specialized engineering knowledge in reusable agent workflows.

In one example, an agentic workflow resolved a highly complex chip-design issue in a day that had been under investigation for weeks. Experiences like that quickly change perceptions. Once engineers see an agent help solve a problem that resisted traditional approaches, adoption accelerates.

This is not about removing engineers from the process. It is about enabling them to work at a higher level and apply their expertise to more problems, more quickly.

Redesigning work for the agentic era

One of Boris’s strongest points was that organizations will not realize the full benefit of AI by simply inserting a new tool into an existing process.

History has shown this before. When computers first entered the workplace, companies that preserved their paper-based processes and used a computer for limited and bespoke tasks saw limited productivity gains. The companies that redesigned their operations around computing saw much greater benefits.

The same principle applies to agentic AI.

Anthropic has placed Claude at the center of a wide range of engineering and business processes, from onboarding and product development to coding, code review, security review, and incident response. Boris described a systematic process of identifying each new bottleneck and then applying Claude to help address it.

That distinction is critical. The objective is not simply to complete today’s tasks faster. The larger opportunity is to redesign how work gets done.

At AMD, we have set an ambitious goal: to use agentic AI to help drive a tenfold improvement in engineering productivity over time. Reaching that goal will require more than deploying tools. It will require changes in workflows, data organization, knowledge capture, leadership practices, and culture.

To learn more about how agentic AI is reshaping chip design at AMD, check out this blog.

As the mechanics of engineering execution become less constrained, the bottleneck also begins to shift. We are increasingly less limited by the speed of execution than by the quality of our ideas, the speed of our decisions and our ability to bring those ideas to market safely.

That puts a premium on a broader set of skills. Engineers will continue to need deep technical expertise, but they will also need curiosity, adaptability, sound judgment, and the ability to work across disciplines. They will need to frame problems well, provide the right context, and evaluate multiple potential solutions.

Agentic tools can also broaden who is able to contribute. A junior engineer, product manager, designer or marketer may be able to test and build an idea that previously required a large, specialized team.

That creates a democratization of opportunity, but it also raises the bar for leadership. Leaders must give employees room to experiment, tolerate intelligent failure, and ensure teams understand the business and technical context surrounding their work. Without that space, people will continue to use new tools in old ways, and the largest gains will remain out of reach.

Some processes are already well suited to agentic AI. Others are not, particularly when data is fragmented, or workflows have not been clearly documented. AI cannot compensate for every underlying process problem. In many cases, it exposes them.

Workflows are not the only thing that has to change. The infrastructure underneath does too, and agentic AI is not only a GPU workload. Every tool call, retrieval step and orchestration decision runs on the CPU, and we see agentic workloads driving roughly 4x the CPU work of a traditional AI query. Scaling agents means scaling both, which is why the full stack matters.

A reinforcing partnership between AMD and Anthropic

This is the context for the strategic partnership AMD and Anthropic announced in July.

Under the agreement, Anthropic plans to deploy up to two gigawatts of AMD Instinct™ MI450 Series GPUs in AMD Helios™ rack-scale solutions, with deployment of the first gigawatt planned for the first half of 2027.

The companies also announced a multiyear engineering collaboration to use Claude to optimize workloads for AMD Instinct™ GPUs and accelerate AMD ROCm™ software development, and AMD will broadly adopt Claude across our engineering and product development teams.

The partnership is powerful because it accelerates innovation in both directions.

AMD provides the open, high-performance and scalable compute foundation needed to support increasingly capable frontier models and agentic. In parallel, Anthropic’s technology helps accelerate our software including ROCm and engineering work required to optimize that foundation and make it easier for developers to build on AMD platforms.

ROCm.ai is another important part of this strategy. Introduced at Advancing AI, this AI-driven developer platform brings AMD expertise directly into leading coding agents, including Claude, enabling developers to understand ROCm natively and build more quickly on AMD platforms. By helping developers install, migrate, troubleshoot and optimize workloads through the tools they increasingly use every day, ROCm.ai helps lower barriers to adoption, improve developer productivity, and create a faster feedback loop to strengthen the AMD software ecosystem.

Together, these efforts create a virtuous cycle: AI helps us build better AI infrastructure, while better infrastructure and software enable the next generation of AI innovation.

They also reflect the full-stack approach we outlined at Advancing AI. The future of AI will not be determined by a single piece of silicon. It will require co-optimization across GPUs, CPUs, networking, rack-scale systems, software, models, and developer tools.

That is the broader focus for AMD: to deliver an open and co-optimized compute ecosystem for the agentic AI era.

Building the agentic future

We are still early in this transition.

Agents will run longer, operate with greater autonomy, and increasingly coordinate other agents. Software teams will be able to take on projects that previously required months or years and complete them in a fraction of the time, and these approaches extend into other domains beyond software.

At the same time, trust, verification, security, and human judgment will become even more important as these systems take on greater responsibility. The organizations that succeed will be those willing to rethink their processes, not simply automate them.

My conversation with Boris reinforced my conviction that agentic AI is not just another productivity tool. It is a new way of organizing work, applying human expertise, and converting ideas into results.

At AMD, we are embracing that shift internally while building the open, full-stack AI platform that will enable developers, enterprises and researchers to bring the agentic AI era to life.

And we are just getting started. 

Watch the full episode of Advanced Insights, featuring Boris Cherny, creator and head of Claude Code at Anthropic.