Welcome back. If you’re overwhelmed by the amount of AI research that’s coming out and just want the highest signal insights, we’ve curated a list of the top 10 AI research papers every engineer should know (along with short summaries). You can access the library here.

Today: Cloudflare releases open-source vibe coding platform, Perplexity gives developers access to its API, and get the latest tutorials, top repos, and trending social posts.

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.

Cloudflare now lets you deploy your own AI vibe coding platform “in one click”: The open-source VibeSDK platform allows users to deploy their own custom AI coding environment in one click, enabling natural language prompts to generate, debug, and deploy full-stack applications in secure, isolated sandboxes on Cloudflare's global network. (You can also watch a tutorial here).

Perplexity releases Search API for developers: The search startup claims that the API provides developers with access to the same infrastructure that powers the company’s public search engine. Perplexity claims the index covers hundreds of billions of webpages and provides structured responses that are ready for use in applications. Read more here.

Tencent unveils open source image model with reasoning powers: The 80 billion parameter HunyuanImage 3.0 image model can reason with world knowledge, understand complex prompts, and generate precise text within images. You can find the model on GitHub or test a version of it here.

Ex-GIPHY VP drops essential document templates for engineering managers: Bjorn Roche, who led GIPHY's engineering team through its $400M Meta acquisition, just published a comprehensive toolkit of management artifacts that every growing engineering org needs.

How to Stay Relevant as an Engineering Leader While Empowering Others: Analysis on 30+ engineering leaders reveals that the most successful managers intentionally create environments where teams take ownership and shine, while leaders stay strategically relevant by shaping vision and building trust.

The 'influence flywheel' that got this engineer to Staff level: Engineering leader Jordan Cutler revealed the counterintuitive operating principles behind his rise to Staff Engineer. His framework centers on what he calls the "influence flywheel" — a self-reinforcing cycle where building influence enables bigger projects.

What actually makes a good software engineer: Engineering leader Candost Dagdeviren has landed on a refreshingly straightforward definition that's gaining traction across dev circles: A good engineer is simply someone you can trust to progress a project and deliver quality solutions consistently while working well with the team.

25 ChatGPT Prompts for Developers: The official OpenAI guide on using ChatGPT to generate code snippets, explain complex concepts, review logic, write documentation, and speed up repetitive tasks so that engineers can focus on shipping high-quality work faster.

The math trick that gave ChatGPT a 4000x memory upgrade: Eight years after the original Transformer paper topped out at 512 tokens, today's LLMs handle 2 million — and the secret wasn't bigger models but smarter position encoding.

A to Z Docker Roadmap with Hands on Labs: A comprehensive Docker learning path just launched on Kimiuz Labs, structured as a visual roadmap that takes developers from basic container commands through production-ready deployments.

GDPval: OpenAI introduces GDPval to measure AI performance on real-world economic tasks across 44 occupations. Frontier models now match or exceed human experts on nearly half of tasks, completing work 100x faster and cheaper, signaling AI's readiness for workplace integration.

Modular Manifolds: Thinking Machines Lab, founded by ex-OpenAI CTO, introduces "modular manifolds"—a framework for constraining neural network weights during training. The research demonstrates Manifold Muon optimizer that keeps weight matrices well-conditioned, improving training stability and enabling automatic learning rate scheduling across network layers.

Metacognitive Reuse: LLMs waste tokens rediscovering identical reasoning steps across problems. Meta's solution: extract common patterns into a "behavior handbook" of procedural knowledge. This achieves comparable accuracy with up to 46% token reduction, fundamentally shifting from "think longer" to "remember how to think."

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

Until next time — The Code team