The Code-Only Agent
Building AI agents with only one tool: code execution. Forces agents to produce reviewable, repeatable code witnesses instead of using traditional tools.
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.
Building AI agents with only one tool: code execution. Forces agents to produce reviewable, repeatable code witnesses instead of using traditional tools.
PHP developer outlines wishlist for 2026 including bundled Composer, simplified extensions, Sets in stdlib, and better version management.
Developer questions what happens to our profession when AI removes the typing constraint, asking whether we solve problems or just translate solutions.
Ugly legacy code often contains undocumented domain knowledge that makes it more valuable than starting fresh.
AI coding agents can rapidly prototype software but still need human expertise for novel ideas, debugging, and production-quality code.
A non-programmer tested AI coding tools to build an app, finding impressive automation capabilities but significant barriers requiring actual programming skills.
This archive is built from SMRTR newsletter stories. Subscribe for hand-picked stories without the extra noise.
Developer shares 10 practical lessons from rebuilding their website in PHP after Hugo upgrade issues, emphasizing familiar tech stacks and focused execution.
Window-art lets Python developers create live visual scenes using desktop windows as interactive canvases for real-time art.
Simple Jekyll implementation serves clean Markdown alongside HTML, making blog content easier for AI agents to parse and consume.
MCP defines five primitives—resources, tools, prompts, sampling, and roots—that separate context, actions, instructions, generation control, and scope boundaries for cleaner agent interactions.
NanoLang combines LLM-friendly syntax with mandatory testing, transpiling to C for native performance while optimizing for AI code generation.
OpenCode with Ollama provides compliant AI coding assistance by running models locally, though performance trails cloud-based alternatives significantly.
Styling radio buttons led to discovering how modern component libraries use 260+ lines of React to recreate native HTML functionality.
A systematic approach to code review using journalist questions, emphasizing test-first reading and strategic timing throughout the workday.
Running multiple coding agents in parallel maximizes productivity by utilizing their autonomous work periods while you focus on planning additional tasks.
Replit's new Mobile Apps platform generates React Native apps from text prompts in minutes, letting users test via QR code on their phones.
Engineers burn out from friction, not code. Context switching, poor prioritization, and communication overhead destroy productivity through systematic workplace inefficiencies.
Fifty years of tools promised to replace developers, from COBOL to AI, but all failed because software complexity requires irreplaceable human reasoning.
Instead of using one AI for everything, developers should use specialized AI agents: Advisors for decisions, Generators for coding speed, and Reviewers for quality.
Step-by-step tutorial showing how to use ReLU activation functions in neural networks to create realistic prediction curves through linear transformations.
Complete guide to React Router v7's data mode featuring protected routes, 2FA implementation, and modern routing patterns.
Meta's StyleX offers an alternative to Tailwind with JavaScript objects instead of utility classes, targeting large-scale applications.
Xenia is a new aesthetically pleasing monospace font for developers with 700+ glyphs and non-ambiguous characters.
Python's yield keyword creates memory-efficient generators that produce data on-demand instead of loading massive lists into RAM.