Welcome back. Big tech has figured it out — the road to AGI runs through your terminal. Cursor just shipped subagents that split your project across parallel workers. Anthropic's Claude Code now runs multi-agent tasks that sync automatically.

Every major lab is betting that whoever cracks coding agents first wins the whole game — and developers are reaping the benefits. Our engineering team compiled 100 Pro Hacks to help you get the most out of Claude Code.

Also: Anthropic’s best practices for using Claude Code, how to launch and generate revenue with B2C apps, and why early-stage startups should avoid managing engineers.

Welcome to The Code. This is a 2x weekly email that cuts through the noise to help devs, engineers, and technical leaders find high-signal news, releases, and resources in 5 minutes or less. You can sign up or share this email here.

Cursor 2.4 brings subagents to speed up AI coding: The AI startup’s latest release introduces subagents — independent workers that handle discrete parts of a task in parallel, each with their own context. Default subagents tackle codebase research, terminal commands, and parallel work streams automatically, while power users can define custom ones. The update also adds in-editor image generation and lets agents ask clarifying questions mid-task without stopping.

Anthropic doubles down on developer tools and AI transparency: Claude Code just got a major upgrade with a new Tasks system, enabling multiple subagents to collaborate simultaneously on complex projects through parallel sessions that automatically sync progress. Meanwhile, the lab also released Claude's behavioral "constitution" under CC0 1.0 — the full spec explaining how the model handles tricky tradeoffs. Developers building their own AI can now freely study and adapt Anthropic's approach.

Liquid AI drops open-source reasoning model that runs on your phone: Two years ago, strong reasoning models required servers. Now Liquid AI's LFM2.5-1.2B-Thinking fits in under 900 MB and runs fully offline. The model uses internal "thinking traces" to reason through problems before outputting concise answers, keeping latency low on-device. It matches Qwen3-1.7B's performance with 40% fewer parameters — a win for developers building math, tool use, and instruction-following apps on edge devices.

Why early-stage startups should avoid managing engineers: A new blog post from an ex-engineering leader at Stripe argues that Seed and Series A founders waste energy trying to "manage" their teams. His advice is clear: hire inherently motivated people, skip the Scrum rituals, and don't bring on engineering managers until you hit 20+ engineers.

What does a staff engineer actually do: GitHub staff engineer Joel Hawksley offers a blueprint for IC growth beyond senior, outlining four key archetypes — Tech Lead, Architect, Solver, and Right Hand — that define the staff role. His key insight for leaders: there must be a clear business need for staff positions, and success comes from helping these ICs "disambiguate, then delegate" complex problems to others.

Palantir outlines five pillars for production-grade AI agents: For engineering leaders deploying AI agents in high-stakes environments, Palantir's new technical deep-dive is a must-read. The company breaks down the security dimensions that matter most — from managing agent memory types (working, episodic, semantic, procedural) to enforcing granular tool access policies at runtime. The bottom line: "There is no reliability without observability," especially with non-deterministic agents.

8 insane OpenCode use cases (save hours): This is the easiest tutorial to get started with OpenCode, a local AI agent that automates coding and system tasks. You’ll learn a simple one-line installation and explore eight powerful use cases — including autonomous documentation and app building — perfect for developers looking to save time immediately.

The longform guide to everything Claude Code: After 10 months of daily use, one developer documented the patterns that separate productive Claude Code sessions from wasteful ones. The key insights: strategic context management prevents "context rot," modular codebases slash token costs, and investing in reusable workflows pays compounding dividends as models improve. Engineering leaders take note — these patterns transfer across AI coding tools, not just Claude.

Gemini CLI: code & create with an open-source agent: Google's new Gemini CLI course is a solid intro to agentic coding assistants. You'll learn how to automate multi-step workflows from your terminal, manage context with gemini.md files, and connect MCP servers to orchestrate across local and cloud environments.

Designing AI-resistant technical evaluations (by Anthropic): This paper discusses how standard technical hiring tests are failing because AI models can now easily beat them. It finds that effective evaluations must switch to unique, unfamiliar puzzles rather than realistic tasks to truly test human reasoning.

Measuring the impact of coding agents on software development: This study explores if autonomous AI coding agents truly improve development speed and quality. It finds that agents only boost speed for teams new to AI, while consistently adding complex, hard-to-maintain code for everyone.

Beyond entangled planning: AI agents often fail at long tasks because they get overwhelmed by mixing up too many details. This research finds that breaking work into isolated steps stops errors from spreading and makes agents much more reliable.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 150K+ 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