Welcome back!

Agentic AI is here, and it's reshaping the tech landscape. A new essay by Sequoia Capital explores how recent AI breakthroughs are moving beyond basic chat responses to advanced, thoughtful problem solving. Could this be the next trillion-dollar disruption? Let’s unpack...

Today’s Summary:

• 🔥 Agentic AI reshapes tech landscape

• 🧬 Google DeepMind CEO wins Nobel for AlphaFold

• 🎞️ Pyramid Flow video AI rivals commercial models

• 🏢 OpenAI expands data center partnerships

• ⚙️ Anthropic unveils batched API for 10k queries

• 🛍 Amazon launches AI-powered shopping guides

• 📸 Imagen 3 available to all Gemini users

• 🌐 Meta AI expands to 21 new countries

• 🛠️ 2 new tools

TOP STORY

Agentic AI reshapes the tech landscape

The Summary: A new Sequoia Capital essay explores how the latest reasoning-oriented AI models could reshape the tech landscape. Rather than relying on rapid pattern matching, AI is moving towards deliberate problem solving. The shift from training-time compute to reply-time compute allows AI to “pause and think” before responding. The future lies in agentic applications, with the potential to transform the software industry and tap into the trillion-dollar services market.

Key details:

• OpenAI's o1 model ”Strawberry” demonstrated reasoning capabilities

• New scaling laws discovered: inference-time compute improves reasoning

• AI companies target service profit pool, selling work outcomes rather than software licenses

• Emerging "cognitive architectures" mimic human thought processes for specific tasks

• Examples of agentic AI apps: Harvey (legal), Glean (work assistant), Factory (software engineering), Abridge (medical scribe)

Why it matters: This evolution in AI capabilities could disrupt entire industries, by automating complex knowledge work and creating new markets. The transition from software-as-a-service to service-as-a-software threatens traditional tech companies and opens opportunities for AI-native startups to gain market share.

GOOGLE

Google DeepMind CEO Demis Hassabis wins Nobel in Chemistry for AI protein breakthroughs

The Summary: Demis Hassabis, CEO of Google DeepMind, and his colleague John Jumper have been awarded the 2024 Nobel Prize in Chemistry for developing AlphaFold2, an AI model that predicts protein structures. This comes one day after Geoffrey Hinton’s Nobel Prize in Physics, marking another recognition for AI research.

Key details:

• AlphaFold, developed by Hassabis and Jumper, can predict the structure of virtually all 200M known proteins

• The AI model has been used by over 2 million people from 190 countries since its release

• David Baker's team created Top7 in 2003, the first protein with a unique structure not found in nature

Why it matters: These consecutive Nobel Prizes recognize the critical role AI plays across scientific disciplines. Alphafold’s breakthroughs have turned tasks that once took decades into matter of hours, with profound implications for medicine, biotechnology, and environmental science. More broadly, AI’s ability to transform fundamental scientific research is now clear.

Scientific background

GUIDE

Pyramid Flow open-source video AI begins to rival commercial models

The Summary: Researchers introduced Pyramid Flow, an open-source model for efficient AI video generation. This new method outperforms current open-source alternatives and is starting to rival commercial models in generating high-quality, long-duration videos. The approach reduce computational demands while maintaining visual fidelity.

Key details:

• Generates 5-10 second videos at 768p resolution and 24 fps

• Achieves top scores in VBench and EvalCrafter benchmarks

• Required only 20.7k A100 GPU hours for training

• Based on SD3 architecture with 2B parameters

• Model available on HuggingFace under MIT license

Why it matters: This advancement could make open-source video generators far more viable by addressing the high compute demands that have previously limited their development. By improving efficiency, Pyramid Flow could help the open-source community close the gap with commercial models.

Research paper - Demo

QUICK NEWS

Quick news

• OpenAI looks beyond Microsoft for AI data centers

• Anthropic Batches API to process up to 10,000 queries at the same time

• Image generation with Imagen 3 now available to all Gemini users

• Meta AI expands to 21 new countries

• Amazon debuts AI-powered Shopping Guides

TOOLS

🥇 New tools

• Miro 2.0 - AI-powered collaboration canvas

• Gradio 5.0 - Build ML apps with a new AI playground to generate apps