TLDR Dev 2026-01-23
How ChatGPT scaled 📈, design engineering 🎨, making videos programmatically 🎥
Clerk MCP Server (Public Beta) (Sponsor)
The Clerk MCP server gives AI coding assistants direct access to current docs, patterns, and best practices. Prompt naturally, get accurate implementations.
Build prompts like:
- "Set up a B2B SaaS with organizations and role-based permissions"
- "Add Clerk middleware to my API"
- "Build a waitlist with user metadata"
Try it today and share your feedback.
Why does SSH send 100 packets per keystroke? (9 minute read)
A developer building a high-performance SSH game found that a single keystroke generated over 100 small packets. The root cause was identified as OpenSSH's keystroke timing obfuscation feature, which sends "chaff" packets to disguise typing patterns. By patching Go's SSH library to disable the `ping@openssh.com` extension server-side, the developer achieved a dramatic >50% reduction in CPU and bandwidth. LLMs assisted in the debugging process.
Scaling PostgreSQL to power 800 million ChatGPT users (14 minute read)
OpenAI scaled PostgreSQL to support 800 million ChatGPT users and millions of queries per second for its predominantly read-heavy workloads. To achieve this, its team offloaded write-heavy operations to sharded systems like Azure Cosmos DB and implemented application and query optimizations. Their architecture uses a single primary Azure PostgreSQL instance and nearly 50 global read replicas.
What is a Design Engineer? (5 minute read)
A design engineer bridges the gap between design and frontend development, creating software that not only works perfectly but also feels great to use. They directly implement designs in code, build and maintain design systems, and add in the critical final details of UI polish, like micro-interactions and animations.
My first year in sales as a technical founder (10 minute read)
A technical founder had a challenging but rewarding first year in sales. High-touch sales are vital for bootstrapped SaaS growth. Their best channel was their LinkedIn B2B sales funnel, with disciplined discovery calls.
Your App Subscription Is Now My Weekend Project (3 minute read)
This dev replaced his subscription apps, like Wispr Flow and Loom, by vibecoding his own personal alternatives. Despite having no prior macOS development experience, he quickly created custom applications that perfectly suited his specific needs for free. However, these apps are not production-ready, but rather perfect for personal projects.
Remotion Agent Skills (Website)
Remotion Agent Skills allow videos to be made programmatically with coding agents like Claude Code.
Townie's back in town! (3 minute read)
Townie v5 is an AI agent powered by Claude 4.5 designed to integrate directly into the browser alongside the code editor. The agent can perform nearly any action a user can, including managing code, files, versions, and databases, allowing for the rapid scaffolding of features or full-stack applications.
The Context Collapse Problem (11 minute read)
AI coding assistants deliver productivity gains on new projects but underperform on legacy codebases, a "context collapse problem" caused by the invisibility of institutional knowledge and historical decisions. Good solutions include treating documentation as infrastructure, incrementally restructuring code for AI legibility, and initially using senior engineers as "human context bridges.”
Keeping 20,000 GPUs healthy (11 minute read)
Modal runs a globally distributed pool of over 20,000 GPUs across multiple cloud providers. Its approach begins with instance type testing and selection, evaluating performance and reliability differences between cloud hyperscalers. This is followed by machine image preparation and automated boot checks to guarantee GPU health before customer workloads begin. Throughout an instance's life, Modal performs continuous passive and active healthchecking.
The most important software engineering news in one daily email
Join 470,000 readers for
one daily email