Weights & Biases

Hey MCP developers! Let’s talk about something broken. 🧠🛠️ Agents are calling tools left and right. But what happens inside those tools? No traces. No visibility. No security. Just a black box. 🕋 The observability gap is real. Let’s fix it together! We believe observability shouldn’t be a bolt-on. It should be a first-class citizen in the agent stack. So we’re launching Observable.Tools — an initiative to bring full-stack tracing to MCP tools using OpenTelemetry. Think: from black boxes → glass boxes. ⎚ To achieve this, we propose combining OTel (OpenTelemetry) into the official MCP protocol via a spec RFC. Combining two vendor-neutral, open protocols to enable easy observability for MCP developers and tool makers (both client and server). 📄 Full proposal details on GitHub: wandb.me/mcp-spec We want to build this ecosystem together. That is why we are extending an invitation to our friends from the observability industry. So… 👀 LangChain, Braintrust, Pydantic, Arize AI, Galileo🔭, AgentOPS & others. Will you join this effort with us, to enable observability for the million incoming MCP developers? Most importantly—this is a call to YOU, the MCP tool developers! ✅ Read the manifesto 💬 Weigh in on the spec RFC 🛠️ Start building tools with observability baked in The agentic future demands transparency. Let’s build it right at https://observable.tools/

Building AI agents shouldn’t mean sacrificing transparency. That’s why we’re excited to team up with deepset to launch a new integration between the deepset AI Platform and W&B Weave! Together we are bringing structured observability to complex, agentic workflows. With this integration, developers can: ✅ Trace every step of their AI pipeline ✅ Debug and fine-tune tools and agents ✅ Monitor performance in real time ✅ Turn black-box systems into explainable, production-ready applications It’s a win for dev velocity, reliability, and trust. Check out the blog post to see how this works in action! 🔗 : https://lnkd.in/ejbCii64 #AIagents #AgenticAI #Observability #LLMops #AIintegration #WeightsAndBiases #Deepset #AIworkflow

Join us and SambaNova Systems to unlock the next level of agentic workflows. We’ll cover: ⚡ How to accelerate inference with RDUs 📊 Evaluating LLMs + agents using Weave 🔍 Monitoring multi-step reasoning 🗓 April 30 ⏰ 8AM PT 📍 Virtual 🔗 Register: https://lnkd.in/g-VhvpPi Perfect for AI devs & ML engineers building enterprise-grade agents.

Mercari, Inc. scaled GenAI beyond prototypes and into real production systems—fast. Their approach? Eval-centric development powered by W&B Weave. Instead of over-investing in prompt engineering or waiting on expensive APIs, they focused on creating high-quality evaluations aligned to real user problems. SMEs and engineers worked in lockstep, using Weave to track, compare, and reproduce 22K+ model runs in just 2 weeks. From seller support tools to internal knowledge systems, this workflow made iteration fast, feedback actionable, and buy-in from leadership possible. The lesson from Mercari US: if you want GenAI to move beyond demos, start with evals. Here’s how they did it: https://lnkd.in/gXWUhezR

🚀 New in W&B Models: Dynamic Grouping for Runs The “Group” property has always been useful for organizing related runs—but until now, it had to be set during logging. That’s changing. You can now move runs between Groups at any time—individually or in bulk. Whether you’re cleaning up a project, reorganizing experiments mid-stream, or retroactively grouping work from multiple teammates, your workspace can now evolve as your workflow does. More flexibility. More control. Less clutter.

W&B’s media panel just got smarter. 🧠 Tracking model outputs by step isn’t always enough. That’s why W&B’s media panel now lets you scroll through images, videos, and other media using any config key—like epoch, global_step, or a custom one. It’s a faster, more intuitive way to evaluate progress and debug models—on your terms.

We’re excited to announce support for GPT 4.1, 4.1 Mini, and Nano in the W&B Weave Playground! Now you can retry production traces, test temperature variance, and compare outputs across models like GPT-4.0, 4.1, and beyond. Get started here: https://lnkd.in/g47wwDt4

Just wrapped an incredible session at #GoogleCloudNext! Our CEO & co-founder, Lukas Biewald, took the stage alongside leaders from Glean, Cresta and Google Cloud to share hard-won lessons and best practices for deploying production-ready AI across the enterprise. From navigating real-world deployment challenges to maximizing ROI and unlocking the next wave of AI-driven innovation—this conversation was packed with insights from those building at scale. Huge thanks to everyone who joined us this morning!

We’re partnering with Google Cloud to shape the future of multi-agent collaboration. As a launch partner for the Agent2Agent (A2A) protocol, Weights & Biases is helping define the open standard for how AI agents discover, negotiate, and interact across boundaries—without needing internal access. A2A supports opaque execution while enabling rich, goal-driven coordination between agents. It’s also fully compatible with MCP: where MCP equips agents with tools, A2A empowers them to find and collaborate with other agents based on advertised capabilities. With W&B Weave, we’re ensuring these systems remain observable, debuggable, and enterprise-ready. Learn more: https://lnkd.in/gaed4ghV

We have an essential session for anyone deploying production-ready AI applications across an enterprise at #GoogleCloudNext! Our CEO & co-founder, Lukas Biewald, will be joined on stage by Glean, Cresta, and Google Cloud for a deep dive into deploying enterprise AI at scale, overcoming challenges, maximizing ROI, and exploring the future of AI-driven innovation. Session Details: 📢 A guide to enterprise AI deployment: Best practices from AI CxOs 📅 Thursday, 04/10 at 8:15am - 9am PT 📍Mandalay Bay Ballroom D 📲 Add it to your agenda now: https://lnkd.in/gRpSZ9Tu #MachineLearning #AI #WeightsAndBiases