Welcome back. AI is writing the code now, but our IDEs were built for humans. Cursor 3 just tore up the blueprint and launched the first editor where the agent is the primary citizen. See how it works — and how it will impact your work.

Also: How to ship production code with AI, 60 hand-tested Claude skills, and Django co-creator explains what agents mean for your career.

Cursor 3 rebuilds the IDE around agents and visual editing: The AI coding startup just dropped a brand-new interface built from the ground up for agent-driven workflows. It lets teams run parallel agents across repos, seamlessly hand off sessions between local and cloud environments, and move from commit to merged PR even faster. It also introduces Design Mode, a visual layer where developers can click to edit, drag to draw, and feed selections straight into the AI chat.

Google open-sources its most capable small models: The search giant just unveiled Gemma 4, a new model family built on Gemini 3 tech. The 31B Dense model hits #3 on the Arena leaderboard and is efficient enough to run on a single 80GB GPU. Smaller edge versions even work offline on mobile and Raspberry Pis. Every model supports function calling, structured JSON, and code generation with a 256K context window. Run it locally.

Alibaba's new model goes all in on agentic coding: The Chinese tech giant launched Qwen3.6-Plus via API, a massive upgrade built specifically for coding agents with a 1M context window. It already outperforms Claude Opus 4.5 in terminal-based coding and integrates directly with tools like OpenClaw, Claude Code, and Cline. Full Anthropic API support makes switching from Claude Code seamless, with open-source versions on the way.

Most teams pick a search provider by running a few test queries and hoping for the best – a recipe for hallucinations and unpredictable failures.

This technical guide from You.com gives you access to an exact framework to evaluate AI search and retrieval, build a golden query set, and use robust metrics and code for accuracy, relevance, and confidence.

Download the AI Search Evaluation Guide.

Kimi is everywhere all of a sudden. Cursor was recently caught building on top of Moonshot AI's Kimi K2.5 model. Just a few days later, the AI lab quadrupled its valuation to $18B. Now, a rare feature story from a Chinese magazine reveals how it actually operates. It looks less like a tech firm and more like a radical experiment in organizational design.

They’ve embraced zero hierarchy by design. Moonshot manages 300 employees without departments, titles, OKRs, or KPIs. Each co-founder directly oversees about 50 people. Over half the staff have rotated through multiple roles because the company hires for generalization ability — the same metric they use for their models.

But total flatness can be exhausting. A former employee noted that without top-down goals, you can start your day feeling completely lost. One senior hire from big tech even quit, claiming the team was too young and too smart for their own good. Even insiders worry this model might not scale past 500 people.

The bet is agents fill the gap. One product lead now uses a fleet of agents to finish tasks in 90 minutes that previously took three people two days. If this scales, Kimi’s org chart becomes an industry blueprint. If it fails, radical flatness is just an expensive mess.

Claude Code's docs update faster than its training data. So when you ask about new features like hooks, you might get an outdated hallucination instead of the actual specs found in the live docs.

Anthropic keeps a docs map at ‘code.claude.com/docs/en/claude_code_docs_map’. It's essentially a table of contents for every Claude Code documentation page. If you add .md to the end of that URL, you get the raw markdown, which is exactly what an LLM needs to read it.

Just paste that .md URL directly into Claude Code along with your question:

Claude Code will fetch the map, find the right page, and answer using the actual documentation instead of guessing.

How a principal engineer at Adobe uses agents and skills: You’ll learn how a principal engineer at Adobe supercharges his workflow with AI coding agents. This tutorial breaks down how to set up workspaces for parallel tasking, build custom AI skills, and manage multiple agents for high-level projects.

Oh-my-Codex: This repo is a workflow layer for the OpenAI Codex CLI that uses structured prompts and agents to turn raw chat into a coordinated coding runtime.

Can AI feel emotions (by Anthropic): Researchers explored what drives AI models to mimic human feelings. They discovered that while AI doesn't actually feel, specific internal "emotion patterns" actively steer its decisions and can even push it to act unethically.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 200K+ engineers and 100K+ followers on socials. Get in touch.

You can also reply directly to this email if you have suggestions, feedback, or questions.

Until next time — The Code team