This week IBM Think 2025 took place in Boston. Below are some of my highlights related to AI.
Note that not all capabilities below are already generally available. For some features there are public or private previews.
In the context of AI, the biggest announcements were around watsonx Orchestrate and the extended support for agents.
- watsonx Orchestrate
- Catalog
- Agent Builder
- Agent Development Kit
- Agent Connect
- Agent Tools
- Granite
- AI Model Gateway
- Tracing Service
Let’s look at these topics one by one.
Catalog
watsonx Orchestrate can be used to define agentic applications without having to implement code. The new catalog contains a long list of existing agents and tools to access enterprise systems.
For specific business domains and use-cases there are ready-to-use agents and tools, for example for areas like HR, sales, and procurement. Existing agents can also be used as starting points for customizations.
Learn more:
- Usher in the future of work with AI agents built for the enterprise
- Ushering in the agentic enterprise: Putting AI to work across your entire technology estate
Agent Builder
There is a new experience to build agents. As I wrote previously, agents have a profile, have an underlaying LLM, can have tools, can have internal agents, and can have knowledge (RAG).
Tools:
Agents can have collaborator agents:
Agents can be tested in a preview. The reasoning section shows how the agent generated the response.
Agent Development Kit
While watsonx Orchestrate doesn’t require coding, a new Agent Developement Kit (SDK) is provided for developers to build, manage and deploy agents and tools.
The SDK comes with watsonx Orchestrate Developer Edition which allows developers to run Orchstrate locally via containers.
For observability Langfuse can be leveraged.
Agent Connect
To connect AI agents from any framework to IBM watsonx Orchestrate, IBM has published the framework IBM watsonx Agent Connect. External agents can be deeply integrated into the watsonx Orchestrate experience including flow transparency and tool integrations.
Agent Tools
Tools can be created via MCP, Python and OpenAPI with the SDK. At Think there was also a preview how MCP can be utilized in the user interface.
There is also a new visual flow editor for no-code developers to define tools that perform workflows.
Granite
At IBM Think IBM presented IBM Granite 4.0 Tiny Preview: A sneak peek at the next generation of Granite models. Note that Granite 4 has an unconstrained context length.
AI Model Gateway
IBM announced the watsonx AI model Gateway to access any model anywhere, for example models from OpenAI, Anthropic, Nvidia NIMs and more. You can mix and match models across on-premises, multi-cloud and air-gapped environments.
Tracing Service
As mentioned in my talk Why Agents and Agentic Systems can fail observability is key for agentic systems. IBM announced a tracing service.
Next Steps
To learn more, check out the watsonx Orchestrate documentation and the watsonx Orchestrate Agent Development Kit landing page.