10 June 2025
At Think 2025, IBM unveiled a powerful suite of upcoming features for watsonx.ai, designed to empower AI builders, data scientists, engineers, and developers to transform innovative ideas into production-ready AI applications that deliver real business value. These enhancements aim to streamline the AI development lifecycle, improve model performance, and provide greater flexibility and control over AI deployments.
One of the most exciting additions is Agent Ops, a new capability that optimizes the end-to-end development and management of AI agents. Agent Ops brings a structured, lifecycle-focused approach to AI agent development. It encompasses everything from building and testing to deploying and managing agents in production.
Watsonx simplifies this process by integrating essential models, tools, frameworks, and evaluation metrics into a unified developer experience. This reduces complexity while preserving the flexibility developers need to innovate. With Agent Ops, teams can build custom agents tailored to specific enterprise needs, optimize their performance, and ensure governance and compliance throughout the development lifecycle. The result is a more efficient path to delivering AI solutions that solve complex business challenges while optimizing for cost and performance.
Another major highlight is the introduction of Model Gateway, a new enterprise-grade solution that provides seamless access to a wide range of leading foundation models through a secure, governed interface. This feature is designed to eliminate vendor lock-in by allowing organizations to integrate IBM’s Granite models alongside frontier models like Anthropic and OpenAI.
Model Gateway offers unparalleled flexibility. It supports practically any model, anywhere, whether hosted on watsonx.ai or in third-party environments. Enterprises can choose the best model for their needs and deploy it securely, thanks to opt-in mechanisms and robust access controls. This capability empowers businesses to maintain control over their AI infrastructure, enhance agility, and optimize costs, all while ensuring resilience and security.
By enabling the use of third-party hosted models, Model Gateway allows developers to build and deploy AI agents as services quickly and efficiently. This flexibility is especially valuable for organizations looking to scale AI initiatives without being tied to a single vendor ecosystem. Model Gateway is available for public preview starting Thursday, 12 June 2025.
Customizing foundation models is a critical step in developing AI solutions that meet specific business requirements. The new model customization features in watsonx.ai make this process more accessible and effective. Developers can fine-tune models using both enterprise data and synthetically generated data, enhancing accuracy and relevance for their unique use cases.
These tools support a range of customization techniques, including tuning and synthetic data generation, enabling teams to optimize model performance while managing costs. Whether you're building a customer service chatbot, a predictive analytics engine, or a domain-specific language model, watsonx.ai provides the capabilities needed to tailor AI to your exact needs.
These newly announced capabilities in watsonx represent a significant leap forward in enterprise AI development. From Agent Ops for streamlined agent lifecycle management, to Model Gateway for flexible and secure model access, and model customization tools for fine-tuning performance, IBM is equipping AI builders with everything they need to succeed.
As organizations continue to explore and expand their AI capabilities, watsonx.ai stands out as a comprehensive platform that simplifies development, enhances governance, and accelerates time-to-value. Whether you're just beginning your AI journey or scaling existing initiatives, these new features can help you achieve your goals more efficiently and effectively.
Statements regarding IBM's future direction and intent are subject to change or withdrawal without notice and represent goals and objectives only.