DeepLearning.AI

Moonshot AI’s new Kimi K2 Thinking and Kimi K2 Thinking Turbo models alternate between cycles of reasoning and tool use, often making hundreds of calls, to outperform other open-weights LLMs on complex, multi-step tasks. Built as trillion-parameter mixture-of-experts models and fine-tuned at INT4 precision, they deliver strong agentic performance while running on lower-cost hardware. Learn more in The Batch: https://hubs.la/Q03VvFRw0

Last week, at AI Dev 25 in New York City, Phillip Snalune, Co-Founder and CEO of Codio, made the case that AI upskilling is the real missing layer in enterprise adoption. “More than half of workers don’t know how to use the technology — and this is about everybody, not just tech teams.” He also underscored the cost mismatch in current training tools and stressed the need for accessible, flexible learning experiences. Phillip also spent time at Codio's demo booth, where he spoke with dozens of developers about AI upskilling.

AI has officially moved from experimentation to engineering and New York made that crystal clear. ⚙️ The AI Dev Conference 2025, hosted by Andrew Ng and DeepLearning.AI, brought together over 1,200 developers and industry leaders to focus on the real work: building reliable, governed, and production-ready AI systems. Bhuvan Reddy Yeturu and Bhaskar Gandavabi represented Fulcrum Digital Inc Digital at the event and the insights they brought back signal a major shift in how the industry is evolving. Across sessions from Vercel, Anthropic, Groq, Arm, LandingAI, AI21 Labs, Databricks, Datadog, and more, one theme dominated: The future won’t be won by bigger models; it’ll be won by better systems. Agentic architectures, multimodal workflows, and efficient on-device intelligence are now at the center of the conversation. Here are the standout signals shaping 2026: 🔹 Engineering > experimentation: Hands-on labs in context engineering and Fintech reliability showed how teams are moving from prototypes to hardened, scalable systems. 🔹 “Small AI” is accelerating: ARM’s efficient neural networks and on-device compute (especially post-Raspberry Pi’s Qualcomm acquisition) point to an era of distributed intelligence at the edge. 🔹 Trust is the new frontier: Panels with Nicholas Thompson, Domyn, and BlackRock reinforced that model integrity, governance, compliance, and privacy are now make-or-break capabilities. 🔹 Agentic platforms are rising fast: Cloud-agnostic toolkits like Amazon’s Strands SDK are attracting serious attention from teams wanting hands-on autonomy without vendor lock-in. 🔹 Developer productivity is exploding: New coding-assistant tools showcased how AI is augmenting developers not replacing them and widening the talent pipeline. And one insight stood above the rest: Every major company across finance, retail, logistics, and tech is now building AI internally with their own teams. They’re not waiting for vendors. They want ready-to-use platforms and agent-driven workflows they can deploy today. For us at Fulcrum Digital, it reinforces exactly where the world is heading: From isolated experiments → to fully integrated, reliable, agentic ecosystems that can be trusted with real business processes. #AIDevConference #DeepLearningAI #AgenticAI #AIEngineering #LLMs #MachineLearning #AIInnovation #FulcrumDigital #AIFuture #EnterpriseAI Which shift do you think will have the biggest impact on enterprise AI in 2026 trust, efficiency, or autonomy? Drop your thoughts below. 👇

Waymo now offers fully autonomous taxi service on freeways in San Francisco, Los Angeles, and Phoenix, the first such driverless service in the United States. Eligible riders can take high-speed, expressway-enabled trips that cut travel time by up to 50 percent. The rollout follows years of testing across public roads, closed courses, and simulations, plus safety collaboration with the California Highway Patrol. Learn more in The Batch: https://hubs.la/Q03Vk7vh0

This week, in The Batch, Andrew Ng reflects on the social energy and technical depth of AI Dev x NYC, and how in-person events spark collaboration, community, and new opportunities in the AI developer ecosystem. Plus: 🚗 Self-driving cars on U.S. freeways 🧠 Moonshot releases Kimi K2 Thinking ⚠️ Anthropic cyberattack report sparks controversy 🔍 Models learn to search their own parameters Read The Batch: https://hubs.la/Q03Vk7ll0

In case you missed it: our PyTorch for Deep Learning Professional Certificate is now available on Coursera. Learn to build, train, and deploy PyTorch models, guided by Laurence Moroney. Start today 👉 https://hubs.la/Q03Vj__V0

What a day! Thank you to everyone who made AI Dev 25 x NYC such a remarkable gathering of builders. The energy in the room, from talks and panels to conversations in the hallways, was unforgettable. If you’d like to revisit key moments of the day, we’ve published a recap. Read it here → https://lnkd.in/eBSmmgnv

🚀 New Course: Semantic Caching for AI Agents Taught by Tyler Hutcherson and Iliya Zhechev from Redis. AI agents often make redundant API calls for questions that mean the same thing. Semantic caching helps your agents recognize when different queries share the same meaning, reducing costs and speeding up responses. In this course, you'll: - Build a semantic cache that reuses responses based on meaning, not exact text matches - Measure cache performance using hit rate, precision, and latency metrics - Enhance accuracy with cross-encoders, LLM validation, and fuzzy matching - Integrate caching into an AI agent that gets faster and more cost-effective over time Start building AI agents that respond faster and cost less to run. 👉 Enroll now: https://hubs.la/Q03T__XB0

Udio settled its lawsuit with Universal Music Group by agreeing to launch a paid platform where fans can generate and remix tracks from UMG artists under artist-set rules (e.g., voice/style use, mashups permitted or not, etc). Artists get paid both for training and for each use, but generated music can only be shared inside the platform (no external downloads or streaming). Learn more in The Batch: https://hubs.la/Q03T-BDw0

Our course "Design, Develop, and Deploy Multi-Agent Systems with CrewAI" is now available on Coursera! Learn from João (Joe) Moura, Co-Founder and CEO of CrewAI, how to build collaborative AI agents that utilize tools, memory, and guardrails to handle real-world workflows at scale. Enroll now: https://hubs.la/Q03TPxxg0