Developer regenerates same app in two languages from shared knowledge files

·1 views

A software developer built a small commerce system by encoding its business logic, architecture decisions, and behavioral contracts as structured files rather than writing code directly. Using that knowledge base, two separate implementations were generated — one in TypeScript and one in Python — without either being derived from the other. Both passed the same 17 contract-based test scenarios, demonstrating that the underlying knowledge, not the code, was the authoritative source. The experiment explores the idea that code may be evolving into a build artifact of documented knowledge, much like compiled binaries are artifacts of source code. The developer acknowledged the analogy is imperfect, notably because AI code generation is non-deterministic unlike a compiler, but argues the approach offers a more reliable way to preserve and transfer system knowledge.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer builds privacy-first emoji mixer that works entirely in the browser

A developer has released a free emoji mixer tool that runs entirely in the browser, requiring no server uploads, sign-ups, or payments. The tool allows users to combine up to four emojis, compositing them with offset, scaling, and rotation before downloading the result as a PNG. Built in roughly 200 lines of JavaScript, it ships all emoji data with the page and functions offline after the first load. The creator noted that most tools marketed as client-side still contact servers for analytics or authentication, and designed this tool to be genuinely local by default. The project is part of a broader collection of around 50 small single-purpose tools the developer has built to handle quick everyday tasks without requiring app installs or account creation.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Academic Coding Falls Short of Industry Demands, Leaving New Developers Underprepared

New software developers frequently struggle when transitioning from academic environments to professional settings, largely because university curricula rarely expose students to large, complex codebases. Real-world projects can span tens of thousands of lines of code, demanding skills in navigation, debugging, and documentation analysis that are seldom taught formally. The absence of adequate documentation in many professional codebases further compounds the challenge, forcing junior developers to infer logic through code analysis alone. Experts point to structured mentorship and hands-on debugging practice as key solutions to help bridge this gap. Addressing this disconnect between academic training and industry expectations is seen as essential for better preparing the next generation of software engineers.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Google Adds Gemini Omni to Vids for Conversational AI Video Editing

Google has integrated Gemini Omni into its Google Vids platform, enabling AI-powered video generation and targeted editing through natural-language conversation. Users can create new clips from text prompts and image references, or apply specific changes to existing footage without rebuilding the entire project from scratch. The model draws on multimodal inputs — including images, text, audio, and video — along with physics-like reasoning to maintain visual coherence during edits such as background replacement, lighting changes, or subject removal. An iterative workflow allows creators to refine selected parts of a video step by step while preserving the broader scene and context. The update also includes optional personal avatars for AI-generated presentations, extending Vids beyond first-draft creation into a more flexible production tool.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Unused MCP server tools silently inflate AI agent costs by up to 28%

A developer benchmarked two otherwise identical AI agents — one connected to a single MCP server and another connected to four — on the same simple question about GitHub repositories. Both agents answered correctly and called only one tool, yet the agent with four MCP servers cost 28% more per query. The cost difference stemmed entirely from extra tool schema definitions being serialized into every API request, adding roughly 4,839 input tokens regardless of whether those tools were ever used. The experiment used an open-source test bench called MAVERIK and ran on Claude Haiku, with results consistent across multiple repetitions. The findings suggest that developers should audit and trim unused MCP server connections to avoid paying a hidden, per-request token tax.

0 comments Read more at DEV Community