OpenAI Function Calling, IBM Watsonx Orchestrate, Amazon AgentCore

The Platform-Native Approach

The orchestration frameworks we have covered so far -- LangChain, LangGraph, Microsoft Agent Framework -- are independent software layers that sit between your application and your LLM providers. But the major AI platform vendors have also built orchestration capabilities directly into their ecosystems. For banking executives, these platform-native options deserve evaluation because they often align with existing vendor relationships, procurement processes, and security certifications.

This unit covers three important vendor-native approaches: OpenAI's function calling and agent tools, IBM watsonx Orchestrate, and Amazon AgentCore.

OpenAI Function Calling

What It Does

OpenAI's function calling is not a full orchestration frameworkOrchestration FrameworkSoftware that coordinates LLMs, tools, and data sources into complex workflows. Frameworks like LangGraph, CrewAI, and vendor toolkits such as the OpenAI Agents SDK and Microsoft Agent Framework manage prompt chains, memory, and tool calling for multi-step AI tasks.See glossary -- it is a model-level capability that allows GPT models to generate structured calls to functions you define. Instead of free text that your application must parse, the model outputs a structured JSON object specifying which function to call and with what arguments. The LLM itself decides when to use tools, which tool to use, and what parameters to pass -- reliably and in a machine-readable format.

How It Works

You define functions using JSON Schema and include them in your APIAPI (Application Programming Interface)A standardized interface that allows software systems to communicate. In AI, APIs let your applications send prompts to a model and receive generated responses programmatically.See glossary request. When the model determines that a function call would help answer the user's question, it returns a structured function call instead of (or alongside) a text response. Your application executes the function, passes the result back to the model, and the model incorporates the result into its response.

For example, a banking assistant might have access to functions like:

• get_account_balance(account_id) -- query the core banking system
• lookup_policy(topic, section) -- search the compliance policy database
• calculate_dti_ratio(income, obligations) -- run a debt-to-income calculation

Banking Relevance

Function calling is the foundation for building AI agentsAgentsAI systems that can autonomously plan and execute multi-step tasks by calling tools, querying data sources, and making decisions without human intervention at each step -- typically within defined permissions and human approval checkpoints.See glossary with OpenAI's models. For banks using OpenAI models through Microsoft Foundry (Azure OpenAI), function calling provides a clean, well-supported mechanism for connecting the model to internal systems. It is simpler than adopting a full orchestration framework when your use cases are tool-calling focused.

Beyond Function Calling: Responses API and Agents SDK

Function calling is only one piece of OpenAI's current agent stack:

• Responses API: OpenAI's agent-ready API, which replaced the older Assistants API
• Agents SDK: an open-source toolkit for orchestrating one or more agents. An April 2026 update added sandboxed execution and memory controls, and the SDK also works with non-OpenAI models

IBM Watsonx Orchestrate

Enterprise AI for Regulated Industries

IBM has a decades-long presence in banking technology, and watsonx Orchestrate reflects that heritage. It is now positioned as an enterprise agent platform -- a "control plane" for building, adding, and governing AI agents across the organization, designed for large organizations with strict governance requirements.

Key Capabilities

Agent catalog and third-party agents. Watsonx Orchestrate offers a catalog of prebuilt agents and can onboard agents built on other frameworks, so a bank can manage agents from several sources in one place. For banks, this means less custom development and a single view of which agents exist.

Conversational orchestration. Users interact with watsonx Orchestrate through natural language, and the platform routes requests to the appropriate agents and tools, coordinates them, and manages the workflow. This makes it accessible to business users, not just developers.

Governance and compliance. IBM has built governance features into the platform from the ground up: audit trails, access controls, monitoring, and explainability. AgentOps evaluation and monitoring is generally available, and an "Agentic Control Plane" announced in July 2026 adds runtime policy enforcement and content guardrails. For banks that apply model-risk disciplines to their AI, these built-in capabilities reduce the compliance burden.

Hybrid cloud support. IBM supports deployment across public cloud, private cloud, and on-premises infrastructure -- critical for banks with data residency requirements or those that cannot host certain workloads in public cloud environments.

Banking Considerations

IBM's existing relationships with many large banks make watsonx Orchestrate a natural conversation, and its emphasis on governance resonates with risk and compliance functions. However, the trade-off is typically less flexibility and innovation speed compared to open-source alternatives.

In April 2026 the Federal Reserve, OCC and FDIC replaced SR 11-7 (and OCC Bulletin 2011-12) with updated model risk management guidance — SR 26-2, OCC Bulletin 2026-13 and FDIC FIL-15-2026. The new guidance is aimed mainly at banks with over $30 billion in assets, and it explicitly places generative and agentic AI outside its scope while the agencies gather input on how banks use AI (the OCC said a request for information is coming). That is not a free pass: examiners can still act on unsafe or unsound practices or violations of law, and most banks continue to apply model-risk disciplines — inventory, validation, monitoring, documentation — to their AI systems.

Amazon AgentCore

Managed Agent Infrastructure

Amazon Bedrock AgentCore, generally available since October 2025, provides managed infrastructure for building and running AI agents. Importantly, it is framework- and model-agnostic: it can run agents written in LangGraph, CrewAI, or other frameworks, using any model.

Key Capabilities

Managed building blocks. AgentCore is a set of services: a Runtime that runs each agent session in isolation, a Gateway that turns existing APIs into MCP tools agents can call, Identity, Memory, Observability, and support for the A2A agent-to-agent standard. AWS runs the infrastructure -- scaling, sessions, and inferenceInferenceThe process of running a trained model to generate predictions or outputs from new input data. Inference cost, latency, and throughput are key factors in enterprise AI deployment.See glossary -- reducing the operational burden on bank teams.

Policy and Evaluations. AgentCore Policy (generally available since March 2026) sets deterministic rules that block an agent's tool calls outside its mandate -- enforced outside the model, so the agent cannot talk its way past them. Think of it as dual control: the agent proposes, but an independent rule decides whether the action is allowed. AgentCore Evaluations (also generally available since March 2026) tests agent quality automatically.

Knowledge bases integration. AgentCore connects to Amazon Bedrock Knowledge Bases, a managed RAG pipeline covering document ingestion, chunking, embedding, and vector storage.

Multi-model support. AgentCore works with models from many providers, through Amazon Bedrock or elsewhere, giving banks flexibility to choose the best model for each use case without vendor lock-in on the model layer.

AWS security integration. AgentCore integrates with AWS Identity and Access Management (IAM), VPC networking, and AWS CloudTrail logging. For banks already running workloads on AWS, this means AI agents inherit the same security controls applied to all other workloads.

Banking Considerations

For institutions already running on AWS, AgentCore is a natural extension of their existing cloud investment, with consistent security, compliance, and operational models -- and its managed nature reduces the scarce AI engineering talent required.

Comparing the Vendor Approaches