Latest story• Feb 17, 2026
Improving Deep Agents with harness engineering
LangChain's coding agent jumped from top 30 to top 5 on Terminal Bench 2.0 using only harness improvements, not model changes.
SMRTR Programming
Programming stories
Get hand-picked daily summaries of the best, most informative programming articles from around the web. Browse recent summaries, then jump to the original source when you want the full article.
Latest stories
We cut Node.js' memory in half (so you don't have to)
V8's pointer compression cuts Node.js memory usage in half with minimal performance overhead, now available through the node-caged Docker image.
How Ralph makes Claude Code actually finish tasks
Agentic AI systems loop indefinitely without clear completion criteria, burning tokens on unnecessary improvements until forced to stop.
WebMCP – a much needed way to make agents play with rather than against the web
WebMCP proposal in Chrome Canary lets developers guide AI agents to content via HTML attributes, replacing inefficient web scraping.
GPT-5.3 Codex vs. Claude Opus 4.6: Which Coding Assistant Should New Developers Choose?
Head-to-head comparison reveals Codex prioritizes speed and execution while Claude emphasizes planning and architectural thinking for coding tasks.
Build a NestJS API for Full-Text Search with OpenSearch
Learn to build a full-text search API with NestJS and OpenSearch, including Docker setup and advanced query features.
SMRTR Programming
Get the next batch of curated stories in your inbox.
This archive is built from SMRTR newsletter stories. Subscribe for hand-picked stories without the extra noise.
More recent stories
Stato captures and transfers AI coding agent expertise across platforms
Stato acts like npm for AI agent expertise, validating and transferring agent knowledge across sessions and platforms with a built-in compiler.
An attempt to make MicroVMs more accessible (SmolVM – Python SDK)
SmolVM provides hardware-virtualized microVMs for safely executing untrusted AI-generated code with sub-second boot times and Python SDK.
My Learning About Password Hashing After Moving Beyond Bcrypt
A developer's journey from bcrypt to Argon2, driven by understanding how modern GPU attacks exploit bcrypt's CPU-only hardness.
A simple CLI to stop copying repos for LLM reference
dotllm manages a system-wide cache of repositories for LLM workflows, creating symbolic links instead of duplicating repos across projects.
The only developer productivity metrics that matter
Most software productivity metrics are easily gamed by developers; only shipping frequency and post-deployment stability truly matter.
Rails 8 Authentication: Why the New Built-in Generator Matters (and What It Means for Devise)
Rails 8's new authentication generator creates a secure, built-in auth system without external gems, reducing reliance on Devise for simpler applications.
TinyDB: A Lightweight JSON Database for Small Projects
TinyDB is a lightweight Python NoSQL database using JSON files and dictionaries, ideal for local scripts and prototypes.
Nike Was Right: The Only Coding Advice You Actually Need
Developer applies Nike's "Just Do It" philosophy to overcome coding procrastination and the trap of waiting to feel ready.
Rolling your own serverless OCR in 40 lines of code
Tutorial shows how to use Modal's serverless GPUs and DeepSeek's OCR model to convert scanned textbooks into searchable markdown for $2.
The Long Tail of LLM-Assisted Decompilation
Deep dive into using Claude and other AI agents for N64 game decompilation, achieving 75% progress before hitting fundamental LLM limitations.
How OpenAI Scaled to 800 Million Users on PostgreSQL
OpenAI scaled ChatGPT to 800M users with PostgreSQL using seven strategies: connection pooling, read replicas, query optimization, caching, sharding, and monitoring.
Building for an audience of one: starting and finishing side projects with AI
Developer used Claude AI to build FastTab, a fast X11 task switcher in Zig, despite having no prior experience with the language.
Build a Real-Time AWS-to-Snowflake Pipeline in 30 Minutes with Python
Tutorial shows building real-time data pipelines with Snowflake's new Snowpipe Streaming Python SDK, AWS Lambda, and Kinesis for sub-second latency.
AI is slowly munching away my passion
A developer mourns how AI has stripped away the determinism, personal identity, and credibility they once found in programming.
How to strace your PHP
Learn to debug PHP applications with strace by tracing system calls to identify blank pages, network timeouts, and file access issues.
Set Up Local LLM Code Completion in VS Code with Ollama
Build a fully offline AI code completion system using Ollama and Continue extension, matching Copilot's productivity without privacy risks.
WebGPU Browser AI: Run LLMs Client-Side, No Backend
WebGPU enables browser-based AI inference with direct GPU compute access, eliminating cloud costs while delivering 3-8x performance over WebGL.
How To Add DRM To Your Backend (easy)
KineMaster blocks modded clients from their asset store by blacklisting app names that include modders' signatures in version strings.