Over the past year, AI agents have moved from an academic novelty to a strategic priority. Agents that can plan, reason, use tools and collaborate with other agents or humans are being integrated into everything from customer service and R&D to enterprise workflows and copilots. According to an analyst report, 96% of enterprises are expanding the use of AI agents.

But here’s the catch: the infrastructure to make these systems interoperable, secure and scalable is still being built. A global race is underway between open source innovators and tech giants to define the rules of this new AI era.

Why standards are critical

Agentic AI isn’t just a smarter chatbot. These systems break down complex goals into actionable steps, call external tools or APIs, maintain context over time and adapt based on feedback. They can work together or with humans, tackling dynamic tasks like managing IT tickets or optimizing supply chains. But to scale this vision, we need shared frameworks for:

• Communication: Agents must speak a common language to collaborate
• Tool integration: Standardized ways to call APIs or services
• Coordination: Orchestrating multi-agent workflows
• Memory: Storing and retrieving context across interactions
• Evaluation: Monitoring and improving agent behavior

Without these, we risk fragmented systems, vendor lock-in and unreliable deployments. With them, we unlock a new software paradigm where AI agents interact as seamlessly as today’s microservices.

Open source vs. Big Tech: Two paths forward

The race to define this future is being driven by two forces: open source communities and cloud giants. Each brings unique strengths and their competition is shaping the agentic landscape.

Open source: Agile and experimental

Visit any AI Engineering team today and you'll likely find them experimenting with open source tools that are evolving at breakneck speed:

• LangChain and LangGraph offer graph-based frameworks for creating stateful, multi-agent workflows with sophisticated control flows
• CrewAI focuses on multi-agent collaboration, assigning roles and enabling inter-agent messaging
• Langflow offers no-code interfaces making agent deployment accessible to non-developers.
• MetaGPT and Open Agents explore emergent behaviors in agent teams
• Emerging standards, like the Multi-Agent Communication Protocol (MCP), aim to provide lightweight rules for agent interactions

Even tech giants are releasing open-source toolkits to accelerate adoption and guide standardization:

Microsoft AutoGen supports multi-agent cooperation, Google’s ADK (Agent Development Kit) offers modular components, Google A2A (Agent-to-Agent Protocol) is attempting to standardize how agents communicate with structured messages and shared goals and OpenAI’s Agent SDK integrates tool-calling with its models.

Big Tech: Enterprise-ready platforms

In parallel, cloud providers are embedding agent capabilities directly into their platforms, focused on production-ready tools:

• AWS Bedrock Agents offers a serverless framework with tool orchestration and memory management, integrating seamlessly with AWS services
• Azure AI agent service provides declarative agents with enterprise-grade security and orchestration
• Google Cloud’s Vertex AI agent builder, tied to its ADK and A2A protocol, emphasizes services for building secure, observable agents with built-in guardrails

These solutions prioritize scalability, governance and compliance critical for regulated industries like finance or healthcare. They’re built to handle complex, secure workloads, but often come with tighter ecosystem integration, which can limit flexibility.

Signs of convergence

Despite the crowded field, common patterns are emerging;

• OpenAPI is becoming the go-to standard for describing tools across frameworks
• JSON-based messaging is gaining traction for clear, interoperable agent communication
• Common memory interfaces are also aligning, with tools like LangChain and Azure AI Search supporting vector databases and hybrid systems

Still, challenges persist. Agent behaviors, message formats and state management vary widely. There’s no universal protocol for what agents can do or how their progress is tracked. Questions around governance, auditability and collaboration protocols remain open.

What’s at stake?

The outcome of this race will shape AI’s future. Will we see a unified Agent Protocol that bridges open source and commercial systems or walled gardens that stifle innovation? How will we audit autonomous agents for compliance and safety? Who will govern their interactions and decisions?

These questions are urgent. A dominant standard could define software development for the next decade. Enterprises need interoperable, secure systems to avoid costly rework. Developers need flexibility to innovate without being tied to a single vendor.

Practical advice for builders

For organizations building agentic systems today, the path forward requires strategic flexibility:

• Prioritize open formats: Use OpenAPI for tools and JSON schemas for messaging to stay flexible
• Abstract your logic: Avoid locking into specific runtimes or providers. Design your agents to run across both open-source and cloud platforms
• Experiment widely: Test frameworks like LangChain, AutoGen or Bedrock to understand trade-offs
• Track emerging protocols: Keep an eye on A2A, MCP and AutoGen AgentSpec, they could set the standard
• Stay flexible: Build flexible, modular architectures and invest in continuous evaluation to adapt quickly as dominant standards emerge

The road to standardization

The road to standardization is still unfolding. No single framework has emerged victorious yet. But one thing is clear: organizations that understand both the technology and strategic implications of these emerging standards will be best positioned to leverage Agentic AI's transformative potential.

At HCLTech, we understand the challenges companies face in such a fragmented Agentic AI landscape. Through a combination of deep domain expertise and strategic investments, we help clients design, build and scale enterprise-grade Agentic AI solutions:

• Modular architectures: We offer open and modular agentic architectures aligned with emerging standards, ensuring flexibility across cloud and hybrid environments
• Strategic partnerships: Collaborations with AWS, Google Cloud and Microsoft Azure enable scalable, secure deployments
• AI and Cloud Native Labs: Our co-innovation spaces allow client's to prototype and benchmark agentic solutions using the latest open source and cloud tools

The Agentic AI revolution is just beginning. By building flexible, future-ready systems, we are empowering enterprises to lead in this transformative era.