TLDR AI⏱ 5 min🛠 builder tools
Cursor 3: A Unified Workspace for Agent-Driven Development
Cursor's redesigned interface focuses on multi-repo workflows and coordination between local and cloud agents for agent-driven development.
Editor's Note
Five different articles today on model coordination and integration architecture. That's not a coincidence.
We've moved past "can AI generate code" to "can AI systems work together reliably." Cursor's multi-repo focus, Weaviate's persistent context experiments, the MCP integration benchmark showing 25-point accuracy gaps — same problem, different angles. Orchestration is where the real work is now.
The MCP benchmark hits hardest. Five integration approaches, most failing silently on real prompts. This mirrors what I see in my own workflows: demo runs flawlessly, production breaks in unpredictable ways. Tool calling still feels like duct tape and prayer.
Simon's scan-for-secrets piece couldn't come at a better time. More people shipping faster means more surface area for accidentally leaked API keys. If you're not scanning before every commit, you're playing with fire.
⚡ Vibe Coding
TLDR AI⏱ 31 min🛠 builder tools
Qwen3.6-Plus: Towards Real World Agents
Qwen3.6-Plus delivers improved multimodal reasoning and a transformative 'vibe coding' experience as a foundation for native multimodal agents.
TLDR AI⏱ 5 min🛠 builder tools
Gemma 4 Open Models
Google DeepMind's new generation of open models optimized for reasoning and agent workflows under Apache 2.0 license.
TLDR AI⏱ 12 min🛠 builder tools
Engram Memory System Deep Dive
Weaviate's Engram shows how persistent context built on vector search improves agent workflows while revealing challenges in reliable tool usage.
TLDR AI⏱ 6 min🛠 builder tools
Open Models Have Crossed a Threshold
Open models like GLM-5 and MiniMax M2.7 now match frontier models for agent tasks at lower cost and latency, making real-world workflows viable.
Simon Willison⏱ 3 min🛠 builder tools
scan-for-secrets: CLI Tool for Finding API Keys in Code
New Python scanning tool helps developers check for accidentally exposed API keys and secrets in their directories before publishing.
Simon Willison⏱ 4 min🛠 builder tools
research-llm-apis: Major LLM Library Redesign in Progress
Simon Willison is redesigning his LLM library to handle new vendor features like server-side tool execution across hundreds of different models.
TLDR AI⏱ 8 min🛠 builder tools
MCP Integration Architecture Determines AI Accuracy
Benchmark of five MCP integration approaches shows 25-point accuracy gap, with most approaches failing silently on real-world prompts.
🧠 The Big Picture
TLDR AI⏱ 4 min🧪 deep analysis
Q1 2026 Timelines Update: Coding Agents Accelerate AI Progress
Faster-than-expected progress in agentic coding over the past few months is moving AI timeline predictions forward, with automated AI R&D coming soon.
LessWrong⏱ 12 min⚖️ AI futures
There Should Be $100M Grants to Automate AI Safety
Apollo Research argues funders should heavily incentivize AI safety work that uses $100M+ in compute budgets on automated AI labor for safety research.
TLDR AI⏱ 6 min🧪 deep analysis
Straight Lines on Graphs: Mental Models for AI Progress
Mental models for accepting that AI progress is remarkably regular and rapid, despite many people's skepticism about exponential improvement data.
LessWrong⏱ 15 min⚖️ AI futures
Democracy Dies With The Rifleman: How Military Technology Shapes Politics
Historical analysis of how military technology changes—from riflemen to drones—fundamentally alter the relationship between citizens and state power.
LessWrong⏱ 8 min🧪 deep analysis
The Bar Is Lower Than You Think
The efficient market hypothesis is wrong, there are no adults in charge, and low-hanging fruit is everywhere if you pay attention to your comparative advantage.
LessWrong⏱ 12 min🧪 deep analysis
Did Anyone Predict the Industrial Revolution?
Historical analysis of why philosophers and intellectuals failed to anticipate the Industrial Revolution and what that means for predicting transformative change.
LessWrong⏱ 15 min🧪 deep analysis
Mean Field Sequence: An Introduction to Understanding Model Internals
First in a series explaining mean field theory as an approach to understanding and interpreting AI model internals, combining explanation with original research.
LessWrong⏱ 6 min🧪 deep analysis
Chicken-Free Egg Whites: Engineering Biology for Food Systems
Engineered yeast now produces realistic vegan egg whites that foam and set like traditional eggs, showing how bioengineering transforms food production.