AI21 Labs

??? New AWS x AI21 Tutorial: Validating E-Commerce Product Descriptions with AI Agents What if your product descriptions could fact-check themselves? ?? In this hands-on tutorial, Yuval Belfer (AI21) and Stuart Clark (Amazon Web Services (AWS) walk through how to build a Product Description Validator using AI21 Labs’ Maestro on Amazon Bedrock. In just 10 minutes, you’ll learn how to: ? Detect and flag inaccurate or inconsistent product info ? Cross-reference descriptions with internal data & trusted sources ? Use Maestro’s Plan & Solve API to automate the process end-to-end ? Implementation code and practical examples that you can apply immediately to your e-commerce applications ?? Watch the full tutorial on YouTube: https://lnkd.in/e83zPyuN #AmazonBedrock #AIAgents #Ecommerce #AI21Maestro #AWSDevelopers

?? Meet the Experts Shaping AI If you're building AI agents and want to hear from experts shaping the field, join us on April 2 for an exclusive event focused on #AIAgents. AI21 Labs, Entree Capital, AWS and NVIDIA are partnering to bring together industry leaders and entrepreneurs for an AI Breakfast Club event you won’t want to miss. Featured speakers include: ?? Ori Goshen – Co-Founder & Co-CEO, AI21 Labs ?? Ronni Zehavi – Co-Founder & CEO, HiBob ?? Nir Sabato – Investor, Entrée Capital ?? Gili Nachum – Principal Gen AI & ML Solutions Architect, Amazon Web Services (AWS) And many more! ?? Limited spots available – secure yours now by registering at the link in comments.

Tomorrow at NVIDIA GTC! ? Join AI21 Labs’ co-founder and co-CEO Ori Goshen as he takes the stage to discuss the transition from reasoning models to AI planning systems. ?? Thursday, March 20 | 9:00 AM PST ?? NVIDIA GTC If you’re attending, let us know in the comments! See you there! ??

?? If you're building AI agents and wondering how to stand out in a crowded landscape, you won’t want to miss the AI Breakfast Club. Join us on April 2 for a morning dedicated to #AIAgents, co-hosted by AI21 Labs, Entrée Capital, Amazon Web Services (AWS) and NVIDIA. Featuring expert speakers: Ronni Zehavi, Ori Goshen and Nir Sabato as well as AI founders Stav Levi Neumark, Benny Porat, Mickey Haslavsky & May Piamenta. Spots are limited, register now ?? https://lu.ma/rrwrhhvh HiBob | Alta | Alta | AI Revenue Workforce | Twine | enso | Vee.com - AI For Good

In today's feature by Pitango, our co-founder & co-CEO Ori Goshen shares the story behind AI21 Labs, our journey and what drives us to build the future of Enterprise AI. From launching Maestro, our AI planning and orchestration system, to continuing to develop the best open model on the market, Jamba 1.6, we're just get started. Thank you Pitango for your support and being part of our journey from the very beginning! ?? Read the full interview. Link in comments. #EnterpriseAI #LLMs #AIInnovation #AccuracyAtScale

? We’re excited to kick things off at NVIDIA GTC 2025 today! Our technical experts will be on-hand at Booth #1839 for personalized 30-minute consultations, helping leaders and decision-makers tackle their biggest AI challenges. Whether it’s private, secure LLM deployment, AI systems for business processes or scaling enterprise AI solutions, we'll help you navigate your unique use cases. What to expect: ?? 1:1 strategy sessions – Get tailored insights on secure LLM deployment, AI system automation and enterprise-grade AI solutions. (Spots are limited—pre-book now!) ?? Live demos of Jamba 1.6, the best open model for private deployment, and preview Maestro, our AI Planning & Orchestration System.? ??? Featured keynote: Join Ori Goshen, co-founder & co-CEO of AI21 Labs, on March 20 at 9:00 AM PST for From Reasoning Models to AI Planning Systems. Will you be there? Let us know in the comments below or sign up to meet our team on site - spaces are limited! https://lnkd.in/eCEqSn8i Looking forward to seeing you there! NVIDIA | Eyal Belfer | Liron Freind Saadon #NVIDIAGTC #EnterpriseAI #LLMs #AIInnovation #AccuracyAtScale

?? New AWS x AI21 Tutorial: Building AI Agents with Maestro Want to see AI21 Labs’ Maestro in action with a real-world use case? In just 13 minutes, this hands-on tutorial with Yuval Belfer (AI21) and Stuart Clark (Amazon Web Services (AWS) walks through how to use Maestro’s Plan & Solve functionality to build an AI-powered Annual Review Writer. You'll learn how to generate comprehensive annual reviews and: ?? Define custom steps with input/output specs ?? Set instructions & constraints? ?? Implementation details of different output types ?? Understand how Maestro’s micro-planning system works behind the scenes ?? Watch the full 13-minute tutorial on YouTube. Link in comments.

ML engineers, AI researchers and AI developers - this one is for you. In their latest conceptual piece, ‘Planning Actions, Not Predicting Tokens’, Barak Lenz, Yonatan Belinkov, Subbarao Kambhampati, Kevin Leyton-Brown and Yoav Shoham argue that Reasoning Models, while an improvement over LLMs, are not the optimal way of exploring the space of possible answers to a given input. Instead, they propose that planning-based alternatives that prioritize structured decision-making over token prediction, offer superior results. The purpose of this paper is not to explore Maestro, AI21 Labs' newly announced AI Planning and Orchestration System, though we’ll share more on that soon, but rather to present how we should be thinking about reasoning and planning in general. At a high level, the paper presents four key concepts: ?? Think actions, not tokens. Reasoning should focus on executing the right action sequences (LLM prompts, web searches, database queries), not just predicting tokens. ?? The space of action sequences is vast—navigate it with planning. Use decision-theoretic planning to weigh cost, accuracy and efficiency. The plan should be the outer loop, while LLMs and other tools serve as the inner loop. ?? At training time, teach the AI system how to plan. At test time, create a specific plan and execute it, interleaving planning and execution. ?? Involve the user throughout the process, especially at test time. The user is both the boss of the system and a resource for providing it with real-time corrective information. Interested in learning more about the inherent limitations of current reasoning models and the benefits of planning-based AI systems? Read the full paper here: https://lnkd.in/esqgpTyS Thank you so much to all of the co-authors for their insights and contributions to this important discussion. It was a pleasure to collaborate with you.

LLMs struggle with complex tasks. AI21 Labs' Maestro - the world's first AI Planning & Orchestration System - doesn't. Check out the below benchmarks to see how Maestro improves outcome accuracy. Sign up for the waitlist and learn more here: https://lnkd.in/e8HAScza #EnterpriseAI #AI21Maestro #AIPlanning #AIOrchestration #HumanX #TrustworthyAI

?? Meet Maestro: The First AI Planning and Orchestration System ?? Today, AI21 Labs announced Maestro - an AI system designed to deliver enterprise-grade AI that organizations can actually trust. Unlike traditional AI approaches that rely on “prompt-and-pray” methods or rigid workflows, Maestro plans, executes and validates. ?? Reliable: Scales inference-time compute, selects the best models and tools and rigorously verifies outputs - ensuring accuracy within defined latency and cost limits. ?? Faster Deployment: Automates AI solution development. Define requirements, connect tools, set a budget. Maestro handles the rest. ?? Adaptive: Learns each enterprise environment, runs offline simulations, predicts success rates and costs and finds the most effective execution strategy for each use case at runtime. ?? Transparent: Provides full visibility with execution traces and validation reports, ensuring clear, controllable and trustworthy AI-driven decisions.and validation reports, ensuring clear, controllable and trustworthy AI-driven decisions. ?? Proven Accuracy Gains. In IFEval, Maestro significantly boosts model accuracy: ? GPT-4o: 85% → 91.9% ? Claude Sonnet 3.5: 88% → 95.2% ? o3-mini: 92% → 95.7% For complex, multi-requirement tasks, Maestro increases accuracy by up to 50% and enables reasoning models like o3-mini to exceed 95% accuracy. ?? Learn more and join the waitlist for early access: https://lnkd.in/e8HAScza #EnterpriseAI #AI21Maestro #AIPlanning #AIOrchestration #HumanX #TrustworthyAI