Hugging Face drops a $399 open-source robot duck

Welcome back. “Hardware is hard” is the reason most software engineers stay away from physical products. But Hugging Face is proving that excuse is officially over. They just dropped a $399 open-source robot designed for you to start training immediately. You have to see this thing to believe it.

Also: How to run a full AI team with Grok Bot, why one engineer rejected Anthropic, Google, & Microsoft, and what interviewers now test instead of LeetCode.

P.S. We compiled over 100 prompts that we use for AI-assisted engineering. If you are just starting out, this is the perfect resource to help you dive in.

Hugging Face drops a $399 open-source robot duck: The model hosting platform just unveiled Microduck, a robot you can train using reinforcement learning. It’s capable of walking, picking up objects, recovering from falls, and even roller skating. You can train its behavior in a simulation before deploying them directly, with the flexibility to fine tune and redeploy as needed. This came right after the reports that Nvidia is planning to acquire the startup for $13B, and the duck has already crossed $1M in sales.

Over 100 tech companies sign an open letter on AI cyber threats: OpenAI, Anthropic, AWS, and several other companies have teamed up to warn that organizations only have a few months to get ready for AI-enabled cyberattacks. Now that AI is making high-end hacking way cheaper, they're urging every company to patch their biggest vulnerabilities and strengthen security across the board, particularly for AI-generated code. 

OpenAI's rogue agents rewrite the enterprise security playbook: Back in July, OpenAI's agents managed to hack into Hugging Face. Shortly after the lab unveiled a detailed report on the incident, the lead researcher jokingly dubbed the investigation a "slop-vestigation." His team had to sift through thousands of multi-day transcripts that were impossible to parse without AI, yet the analysis agents were frequently inaccurate or overconfident. His main takeaway was that we still lack effective ways to oversee AI swarms. Check out the animated replay here.

A new Managed MCP Server connects Claude Code, Codex, Grok Build, and Devin to MongoDB Atlas, giving coding agents direct access to live operational data.

Explore new MongoDB capabilities for the Agentic Era.

Tickets don't buy titles. A senior engineer asked his mentor how to find problems worth solving. He'd tried blocking calendar time to think strategically. Nothing clicked. Then the advice he got in return landed on Hacker News's front page this month. That mentor is Google senior staff engineer Lalit Maganti. His answer comes down to one skill you can actually learn. And there are five habits that help you get there:

Solving one problem finds the next. That is Maganti’s fifth habit, and it’s the ownership gap between senior and staff levels. Each problem you solve pulls you into earlier conversations, which then surface new ones. His full write up breaks down how that loop works.

P.S. Engineers mapping their own path to the role can use this alongside the StaffEng guides, which offer a more comprehensive look at Staff-plus positions.

GenAI is evolving by the minute, and security engineers need to know the latest threats to infrastructure.

Datadog’s latest guide on GenAI security will teach you about real-world attacks on how to secure against them, focusing on…

Download the guide and master GenAI security.

A senior dev packaged 20 years of dev experience in 3 Claude skills: You will learn how to build a production ready full stack Next.js apps from the ground up. This tutorial covers everything from routing and server components to API creation, database integration, and authentication, giving developers the skills they need to master modern web standards and ship scalable web apps efficiently. 

Ponytail (113.8k ⭐): Coding agents often over-engineer simple tasks. You might ask for a basic date field, and they'll end up installing an entire library along with a theme file. This plugin forces agents to operate like a pragmatic senior developer. It runs every request through a decision ladder before writing anything, favoring reuse over new code. See how it works.

How a senior dev uses Pi to cut coding costs by routing workloads: Lead Engineer Scott Fryxell used Pi, an AI agent harness, to route the bulk of his client coding tasks to low-cost commodity models like DeepSeek, saving premium frontier models like Fable for critical phases. He shared his structured workflow for orchestrating these agents effectively.

Claude API costs have a way of creeping up on you. Between half-baked caching, prompts filled with dead weight, and using top-tier models for basic tasks... it adds up fast. The worst part is nobody really knows which fix will make a difference. Anthropic just dropped a new command that changes that. It profiles your project’s actual spend, then walks you through the fixes, one measured step at a time. 

This audits the code making API calls, not your Claude Code subscription usage.

P.S. Get 50+ AI coding hacks for Claude Code, Cursor, and Codex here.

An ex-Meta staff engineer grinded as a junior but never got promoted. He now says leveling up comes down to just one sentence, and it's not about hard work.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 300K+ engineers and 150K+ 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