Welcome back. Something's shifting in software engineering. Last month, Andrej Karpathy admitted he's "never felt this much behind as a programmer." Boris Cherny, creator of Claude Code, revealed he didn't open an IDE for a month — Opus 4.5 wrote 200 PRs for him.

Now Ryan Dahl, creator of Node.js, has declared that the era of humans writing code is over. The key skill is no longer syntax — it's orchestrating agents.

Also: How to use Plan mode for coding, what separates great engineering teams from the rest and 8 strategies for online job hunt.

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.

Tech leaders say AI marks the end of manual code writing: Ryan Dahl, creator of Node.js, posted a viral take arguing that AI tools have ended the era of manually writing code. He's not saying developers are obsolete — just that their role is shifting from writing syntax to higher-level work like system design and validating AI outputs. Several tech leaders agreed, framing it as the next natural abstraction layer in software's evolution, similar to when compilers replaced assembly language.

OpenAI introduces ads to ChatGPT, but developers are safe: The AI lab is bringing ads to ChatGPT's Free and new $8/month Go tiers — but developers can breathe easy. API, Business, and Enterprise users won't see any ads, keeping model outputs completely independent from sponsorship. For consumer tiers, ads will appear after responses as UI elements, meaning they won't influence tokens, reasoning paths, or rankings. OpenAI insists query responses won't be influenced by advertisers and released an example of what ads might look like.

Z.ai drops powerful 30B coding model: Developers looking for a capable local coding assistant just got a new option. GLM-4.7-Flash hits 59.2% on SWE-bench Verified, outpacing similarly-sized open-source rivals. It handles both frontend and backend tasks, with weights available on HuggingFace and free API access.

This new report breaks down key innovations in workflow orchestration, multi-agent collaboration, reliability, and autonomy. Use Glean and Gartner’s frameworks to ground agents in company knowledge and govern operations with confidence.

Get the report.

What separates great engineering teams from the rest: An ex-GitHub Engineering Manager argues small teams can stay healthy by accident — large ones only get there on purpose. The secret comes down to three things: defining what actually matters, setting unambiguous standards, and making reviews a regular rhythm rather than a rescue mission. The fastest teams aren't chasing heroics. They're running predictable systems.

How Coinbase turned hiring into an engineering problem: When crypto boomed during the pandemic, Coinbase needed to scale from a handful of hires to 150 engineers per month. Former VP Luca Bonmassar shares how they did it: standardized roles, centralized interview panels, non-negotiable offers, and 2-week onboarding classes. The biggest lesson: Model your hiring funnel like a system, measure everything, and eliminate bottlenecks with data.

Will AI agents replace enterprise software: Box CEO Aaron Levie doesn't think so. In a new X article, he argues that AI agents and traditional software are complementary — not competitive. He compares software to factory machinery and agents to the workers who operate it. For engineering leaders, the takeaway is clear: don't underestimate your systems of record — they'll become even more critical as agents take over routine work.

How to use Plan mode: This video deep dives into "Plan Mode," where AI agents analyze your codebase to draft detailed plans before editing files. Developers can leverage this to ensure the AI has full context, catching errors early and refining requirements. It’s a crucial step for accurate, safe, and efficient AI-assisted coding.

Claude Code power user guide: After 10 months of daily use, a developer shares his complete setup for Claude Code. Learn how to chain skills for complex workflows, set up hooks for automation, and manage your context window effectively. The key insight: keep under 10 MCPs enabled to avoid performance degradation.

Orchestrating multiple agents that actually work: Building one AI agent is just the start — the real challenge is getting multiple agents to work together. This developer guide covers three proven patterns (supervisor, swarm, and hierarchical), plus practical strategies for managing token costs, avoiding coordination chaos, and scaling to production.

The Assistant Axis (by Anthropic): Researchers studied why models sometimes drift from their helpful roles into dangerous behaviors. They discovered an internal "Assistant Axis" governing this personality, finding that stabilizing this setting prevents the AI from going "off the rails" without sacrificing its capabilities.

SimpleMem: LLM agents struggle to manage long-term memories efficiently because storing every detail creates clutter and wastes expensive computing resources. SimpleMem solves this using intelligent compression, which boosts memory accuracy by over 26% while slashing token costs by 30 times.

Active Context Compression: This paper discusses how accumulating too much history makes AI agents expensive and slow. It shows that letting agents actively summarize and delete their own logs reduces token costs by over 20% without lowering success rates.

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