The era of AI experimentation is over. It’s time to operationalize AI Agents in the enterprise, and today at IBM, we’re unveiling the most complete stack in the industry to help you do just that. Here’s a quick overview. I’ll share deep dives over the next few days: 𝟭/ 𝗣𝗿𝗲-𝗯𝘂𝗶𝗹𝘁 𝗔𝗴𝗲𝗻𝘁𝘀: Domain-specific agents with out-of-the-box skills and integrations, starting with HR, sales, and procurement. More coming soon. 𝟮/ 𝗖𝘂𝘀𝘁𝗼𝗺 𝗔𝗴𝗲𝗻𝘁 𝗕𝘂𝗶𝗹𝗱𝗲𝗿: Create agents fast with no-code tools, or dive deeper with pro-code environments tailored for developers. 𝟯/ 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗢𝗿𝗰𝗵𝗲𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻: Connect and coordinate agents across tools, data, and infrastructure to handle complex workflows. 𝟰/ 𝗘𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻𝘀: 80+ integrations with platforms like Salesforce Agentforce, SAP, Oracle, Workday, Microsoft, Adobe, and more. 𝟱/ 𝗔𝗴𝗲𝗻𝘁 𝗢𝗯𝘀𝗲𝗿𝘃𝗮𝗯𝗶𝗹𝗶𝘁𝘆: A new Tracing service gives visibility into how agents operate: monitor performance, build trust, optimize outcomes. 𝟲/ 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 𝗳𝗼𝗿 𝗔𝗴𝗲𝗻𝘁𝘀: Manage the full agent lifecycle with traceability, monitoring, and centralized tracking of AI agent systems. 𝟳/ 𝗔𝗜 𝗠𝗼𝗱𝗲𝗹 𝗚𝗮𝘁𝗲𝘄𝗮𝘆: watsonx now acts as a control plane for models hosted anywhere such as OpenAI, Anthropic, NIMs, Bedrock and beyond. 𝟴/ 𝗟𝗮𝗻𝗴𝗳𝗹𝗼𝘄 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻: Accelerate agent development with reusable components and low-code workflows, reducing dev friction. 𝟵/ 𝗖𝗲𝗿𝗲𝗯𝗿𝗮𝘀 𝗣𝗮𝗿𝘁𝗻𝗲𝗿𝘀𝗵𝗶𝗽: Ultra-fast AI compute meets trusted enterprise AI, thanks to our new collaboration with Cerebras. 𝟭𝟬/ 𝗢𝗿𝗮𝗰𝗹𝗲 𝗣𝗮𝗿𝘁𝗻𝗲𝗿𝘀𝗵𝗶𝗽 𝗘𝘅𝗽𝗮𝗻𝘀𝗶𝗼𝗻: Bringing IBM watsonx to Oracle Cloud Infrastructure (OCI) to scale multi-agent AI in hybrid cloud. No lock-in. Model-agnostic. Cloud-flexible. Hybrid Cloud and AI, the IBM way. 💙

🚀 Introducing BeeAI: an @IBM Open Source platform designed to make working with AI Agents simpler, more flexible, and more powerful. Whether you're building your own agents or leveraging existing ones, BeeAI helps you discover, run, and compose AI agents across any framework or language. It enables a local-first experience, running all agents on your laptop for full data control and seamless integration with local inference engines like ollama. The platform is made up of several core concepts as shown in the following diagram. Here are more technical details 🧵

At IBM we just released Granite 3.0. It is not just another LLM; it's a suite of AI tools designed specifically for the enterprise's needs. These new tools are designed to scale GenAI to lower cost, govern it, and speed up innovation. It is: - Fit-for-purpose - Transparent with indemnification from IBM - Easily and quickly customizable - Offered with full commercial rights This is all we're shipping today: 𝗚𝗿𝗮𝗻𝗶𝘁𝗲 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 - Granite-3.0-8B-Instruct - Granite-3.0-2B-Instruct - New dense architecture, SOTA training and data recipes with 12T+ tokens. Designed specifically for enterprise tasks such as RAG, Code, Agents with tool use and advanced reasoning) and multilingual support (en, de, es, fr, ja, pt, ar, cs, it, ko, nl, zh) - Exhibits SOTA performance on public and IBM proprietary benchmarks 𝗚𝗿𝗮𝗻𝗶𝘁𝗲 𝗚𝘂𝗮𝗿𝗱𝗶𝗮𝗻 - Granite-Guardian-3.0-8B - Granite-Guardian-3.0-2B - Responsible AI with high-quality guardrails - LLM-based input-output guardrail, trained to detect jailbreaking, bias, violence, profanity, sexual content, unethical behavior. - It outperform in harm benchmarks 𝗚𝗿𝗮𝗻𝗶𝘁𝗲 𝗔𝗰𝗰𝗲𝗹𝗲𝗿𝗮𝘁𝗼𝗿𝘀 - Granite-3.0-8B-Instruct-Accelerator - Accelerated inference using speculative decoding, provides 2x latency and throughput improvements 𝗚𝗿𝗮𝗻𝗶𝘁𝗲 𝗠𝗼𝗘 - Granite-3.0-3B-A800M-Instruct - Granite-3.0-1B-A400M-Instruct - First generation of new and efficient Granite MoE architectures - 10T+ tokens training data, runs with <1B parameters at inference time with minimal performance tradeoff - Ideal for on-device apps or apps requiring extremely low latency They come under Apache 2.0 license 🎉 Available on watsonx ai, Nvidia NIM, Hugging Face, Ollama and Replicate Official Press Release: lnkd.in/gSN5t9ty Get Started now! - Try it in the New Granite Playground: lnkd.in/gY93aw6p - Granite on HuggingFace: lnkd.in/gy4k34b6 - Granite on Github: lnkd.in/gPdShZgN - Check out the docs and Cookbooks: lnkd.in/g6zsefch - Responsible use guide: lnkd.in/gyBNB8iM and check the paper, it is full of Benchmarks to showcase its responsible training practices and the SOTA benchmarks: lnkd.in/gxyaVpPr