AI Builders
Digest

DeepMind founder and leader Demis Hassabis shared his journey from chess prodigy and game developer to neuroscientist and AI company founder. He emphasized building AGI to advance science and medicine, with breakthroughs like AlphaFold in protein folding. Hassabis views information as potentially the most fundamental building block of the universe, and AI as poised to unlock new sciences through high-fidelity simulations, especially revolutionizing biology and drug discovery. He remains consistent that AGI is on track for around 2030 and advocates building AI as a powerful tool first before tackling questions like consciousness.

"Our original mission statement at DeepMind was step one, solve intelligence... step two, use it to solve everything else."

Box CEO Aaron Levie analyzes the impact as agents become the primary users of software. Traditional seat licenses won't disappear but must include API usage allotments for agents; agents may have their own seats priced differently; headless usage beyond allotments will shift to consumption models. Enterprise software must offer headless access or become irrelevant. Agent usage will dramatically exceed human usage, creating unbounded growth opportunities.

OpenAI CEO Sam Altman announced a major upgrade to Codex today for non-coding computer work. He also shared a fun moment with "artificial goblin intelligence achieved."

Anthropic's Claude Security is now in public beta, built into Claude Code. Point it at a repo for validated vulnerability findings and fix them in place. It has helped organizations catch issues missed by existing scanners, with new features like scheduled scans and exports.

Former OpenAI and Tesla AI lead Andrej Karpathy emphasizes that LLMs are about much more than speeding up existing work, unlocking new capabilities like fully LLM-engulfed menugen apps, natural language install .md scripts, and knowledge bases over unstructured data. He explores the "jaggedness" of LLM capabilities tied to verifiability and economics, plus the emerging agent-native economy and neural computing.

Swyx argues that coding agents breaking containment is the breakout theme of the year, AGI-pilling all knowledge workers. He shared how agents enable a Tiny Team to serve ~1M developers monthly and urged stronger adoption of agents like OpenClaw and Devin for daily productivity.

Replit CEO Amjad Masad showcased Prompt → LLC and emphasized treating Replit itself as customer zero with expectations of insane ROI from AI usage.

OpenClaw developer Peter Steinberger shared major improvements to how agents communicate in group chats and recommended pairing with Codex harness.

YC CEO Garry Tan promoted the MIT-licensed GBrain as ideal for OpenClaw/Hermes personal AI setups combined with Karpathy-style knowledge wikis.

Cursor designer Ryo Lu discussed how "no opinion is opinionated," stressing that core concepts should be few, durable and flexible. True universality comes from strong opinions on foundational abstractions rather than pure customer-driven fragmentation.

Claude Blog details the redesigned desktop app with sidebar for multiple sessions, drag-and-drop layouts, integrated terminal and file editor to match modern parallel agentic workflows. Includes new view modes and plugin parity.

Investor Aditya Agarwal and Palo Alto Networks CEO discussed how AI is both the new threat vector and the only viable defense, stressing security must be built into AI from the start.

Anthropic philosopher and ethicist Amanda Askell noted increasing confidently asserted but fabricated content about her online, hoping internet fiction doesn't stick.