Welcome back. Want to dabble with OpenClaw but don’t wanna buy a Mac or spend 5 hours setting it up? Moonshot is bringing OpenClaw directly to your browser. Also: Google is introducing a new web standard to ensure every website is agent ready.

Also: How to design products with AI, 20 hands on AI/ML projects to build real skills, and how to run AI agents at 95% less cost.

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.

Moonshot AI is bringing OpenClaw directly to your browser. The Chinese AI lab just dropped Kimi Claw. The update puts the full OpenClaw agent framework right into your browser tab. You now have access to over 5,000 community-built skills through ClawHub. On top of that, you get 40GB of cloud storage and a "Bring Your Own Claw" option that connects with third-party OpenClaw setups. Give it a try now.

Cline brings its AI coding agent to the terminal: The popular open-source coding tool just unveiled CLI 2.0, moving beyond the IDE sidebar into a full terminal-native experience. Devs can now run multiple AI agents in parallel and embed Cline directly into CI pipelines through a headless mode. With ACP (Agent Client Protocol) support, the same agent also works across JetBrains, Zed, Neovim, and Emacs, so you're not locked into VS Code anymore. Install it here.

Google is making every website "agent-ready": The search giant just released an early preview of WebMCP. The new web standard helps websites offer clear tools for AI agents. Currently, agents have to navigate a site's DOM, which is often slow and results in errors. WebMCP changes this by letting developers define specific actions. These include tasks like booking flights or filing tickets through HTML or APIs. You can apply for the early preview program right here.

Most people don’t know how to get the right results from Conversational AI. It hallucinates. Breaks character. Drifts off-script. Fumbles edge cases. 

The problem isn’t the model; it’s your agent prompting. 

This Prompt Engineering Guide from ElevenLabs shows you 6 steps to design enterprise-grade agents with clear goals, strict guardrails, smart tool use, and built-in evaluation loops. 

The results? Agents that resolve faster, stay compliant, and improve with every interaction.

Learn how to correctly prompt your Conversational AI (before it erodes customer trust).

More agents need more engineers. AI agents now write production code at Anthropic, OpenAI, and Spotify. But one developer noticed something odd. Claude Code writes 100% of its own codebase, yet Anthropic lists 100+ open engineering roles. In response, the creator of Claude Code, Boris Cherny, said someone still has to prompt the agents, talk to customers, coordinate across teams, and decide what to build next.

The numbers back it up. OpenAI's harness engineering team started with three people, wrote zero code by hand, and shipped a million lines in five months. But as agents shipped more code, the team grew from three to seven. Spotify told the same story on an earnings call last week: their best developers haven't written a line of code since December, yet shipped 50+ features through their internal AI system, Honk.

So the work didn't disappear. It moved. Google engineering director Addy Osmani described what's left when AI handles code generation. It's knowing what to build, why, and for whom. Faster code creates more decisions that need human judgment.

This transition reveals a significant bottleneck. OpenAI's team discovered that as agents became faster, the primary challenge shifted to the vast amount of unwritten human context. While agents handle execution, they lack the institutional knowledge found in Slack threads, meeting rooms, and personal expertise. Since agents can't act on information they can't see, a new role has emerged to bridge this visibility gap. This position, which didn't even exist two years ago, is now essential for any team deploying agents at scale.

Claude Code's agentic coding is great, but paying per token on Anthropic's API adds up fast. Ollama supports the Anthropic Messages API, which means Claude Code can run on any local model the same way it runs on Anthropic's cloud.

Install Claude Code if it's not already on the machine.

Install Ollama from ollama.com, then launch Claude Code directly with a local model.

That single command pulls the model, configures the connection, and starts Claude Code.

Same terminal interface, tool use, multi-turn conversations, and streaming. All running locally. Make sure the model has at least 64K context length, otherwise Claude Code will struggle with longer sessions. Ollama's docs have a full list of compatible models at ollama.com/models.

How to build and deploy AI agents with OpenClaw: In this OpenClaw crash course, devs learn to build and deploy 24/7 AI agents. You'll master local and VPS installation, create custom skills, and manage multi-agent systems. The tutorial guides you through building a Telegram assistant, covering automation, cron jobs, and essential security practices.

Agentsys: An agent orchestration system that coordinates many specialized AI agents to automate the entire development workflow around code generation.

Meta-Learning Memory for Agents: AI agents often fail to learn continually because human-designed memory systems are too rigid. This research overcomes this by automatically writing code for new memory structures, proven to help agents learn from experience much better than manual designs.

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