Partner POV | The Importance of Open Ecosystems for Agentic AI Deployment

Written and provided by: AMD

The arrival of agents -- defined as bots that can take independent action across multiple applications or services to accomplish a goal -- has raised customer expectations and challenged businesses to consider how agentic AI will interact with existing software and already-defined workflows. AI agents rely on large language models just as chatbots do, but agentic systems are designed to use software tools and data sources that generative AI chatbots often cannot reach. Those resources can include customer records, internal libraries, ticketing systems, code repositories and business applications. 

Agentic AI requires an open ecosystem

Agents need this access as well as broad software interoperability to carry out their tasks, but companies attempting to scale from generative to agentic AI may find themselves stymied by vendor lock-in and proprietary software solutions that only work well with a handful of other applications. The same vertical integration that can make initial deployment easier can become a liability when a company expands its AI usage and the suite of tools it supports.

A recent article in the Wall Street Journal discusses the value of open AI ecosystems and the benefits of solutions that give businesses room to experiment. Currently, many agents may work well within specific software frameworks but can struggle when asked to interface with other applications. Maribel Lopez, principal analyst at Lopez Research, told the publication that deployment flexibility is a critical component of successful AI rollouts over the long term.

"There's probably never been a tech stack we've dealt with in our lifetime that has required innovation this rapidly," Lopez told the WSJ. "AI is not a set-it-and-forget-it technology—you always have to be building on it. There can be a lot of value in having access to that new, fresh, constantly evolving tool set that the open-source community offers."

AMD's open approach to Enterprise AI

AMD supports the industry's need for rapid, continuous innovation with a comprehensive suite of end-to-end AI solutions that spans embedded FPGAs, AMD EPYC server CPUs, mobile workstations and AMD Instinct GPU accelerators. This breadth gives customers more room to match AI workloads to the power and cooling limits of the environments where these workloads will run. AMD ROCm and HIP support industry-standard frameworks like PyTorch and TensorFlow, as well as the ONNX runtime, allowing models to be evaluated, ported and deployed across environments without retraining or reformatting. This flexibility can help organizations respond swiftly to conditions today, without foreclosing strategic options whose importance may only become apparent months or years down the line.

Additionally, AMD provides containerized services, such as AMD Inference Microservices (AIMs), to manage AI inference workloads on AMD Instinct GPUs. AIMs provide Enterprises with an "easy" button for inference, taking away all the complexity while providing service-level assurance when deployed on specific AMD hardware configurations. These microservices are part of the open-sourced AMD Enterprise Reference Stack and are designed to interface with a company's existing environments and toolsets through open and closed source platform partners such as Red Hat, SUSE, Canonical, VMware and Nutanix. AMD's goal is to enable agents and data connector functionality across structured and unstructured databases, spreadsheets and SaaS products. Achieving this requires an architecture that does not restrict access to disparate data sources or associated applications.

Flexibility and freedom of choice will only become more important as agentic AI matures. Early agent tests may automate a narrow task inside a single application, but the advantages of full production deployment will require agents to access multiple data sources, launch appropriate tools and return results in formats other software can use. Companies that prioritize agents capable of connecting to existing employee workflows and systems will be in the best position to realize the benefits agentic AI provides. The open ecosystem approach, championed by AMD, gives enterprises a path for scaling AI throughout their organizations without limiting available software options or forcing every future decision through a single vendor software stack.