Published on December 23, 2025 at 18:01 CET (UTC+1)
Test, don't (just) verify (102 points by alpaylan)
The article argues that AI is bringing formal verification—the mathematical proof of software correctness—into mainstream software engineering. It highlights how AI is solving complex challenges like the lack of formal specifications and is achieving remarkable results in competitions like the IMO. The core proposal is to move beyond just testing software to actively proving its correctness using AI-assisted tools like proof assistants.
Adobe Photoshop 1.0 Source Code (1990) (325 points by tosh)
This post from the Computer History Museum details the release of the original Adobe Photoshop 1.0 source code from 1990. It recounts the program's origin as a personal project by brothers Thomas and John Knoll, its initial bundling with a slide scanner, and its eventual acquisition and launch by Adobe. The article serves as a historical record of one of the most influential software applications ever created.
Instant database clones with PostgreSQL 18 (258 points by radimm)
The article explores a new, efficient method for creating instant database clones in PostgreSQL 18. It explains how to leverage the database templating system, combined with modern file system features like reflinks on XFS, to create copies of large databases with minimal storage overhead and near-zero time. This is positioned as a solution for testing, development, and safe data manipulation.
Ryanair fined €256M over ‘abusive strategy’ to limit ticket sales by OTAs (125 points by aquir)
Italy's competition authority has fined Ryanair €256 million for abusing its dominant market position. The airline was found to have deliberately implemented technical obstacles to hinder online travel agencies (OTAs) from selling its tickets, aiming to force all sales through its own website. The fine covers conduct from April 2023 to at least April 2025.
Executorch: On-device AI across mobile, embedded and edge for PyTorch (62 points by klaussilveira)
Executorch is an open-source framework from PyTorch for deploying AI models on mobile, embedded, and edge devices. It enables efficient on-device inference, allowing models to run directly on end-user hardware without needing a cloud connection. The project provides tools to prepare and execute PyTorch models in resource-constrained environments.
Font with Built-In Syntax Highlighting (2024) (98 points by california-og)
This blog post introduces an innovative concept: a font with built-in syntax highlighting using OpenType features. The goal is to enable syntax-colored code snippets on hand-coded websites without relying on external JavaScript libraries. The font uses character substitution to apply different glyphs/colors based on context, simplifying website complexity and bloat.
Show HN: Yapi – FOSS terminal API client for power users (13 points by jamiepond)
Yapi is a free, open-source (FOSS) command-line API client designed for power users. It allows developers to test APIs (HTTP, gRPC, TCP, GraphQL) directly from the terminal using declarative YAML configuration files. It includes features for environment variables, automated response assertions, and integrates with tools like jq for parsing.
Show HN: CineCLI – Browse and torrent movies directly from your terminal (242 points by samsep10l)
CineCLI is a cross-platform, command-line application built with Python that allows users to browse, search, and torrent movies directly from their terminal. It pulls data from the YTS movie database, displays detailed information, and can launch magnet links into the user's default torrent client. It emphasizes a fast, minimal, and text-based interface.
Carnap – A formal logic framework for Haskell (81 points by ravenical)
Carnap is a free and open-source framework built with Haskell for teaching and studying formal logic. It is used as an educational platform at numerous universities worldwide, providing students with interactive tools for completing logic exercises. The website serves as a portal for students and educators interested in the system.
10 years bootstrapped: €6.5M revenue with a team of 13 (193 points by steffoz)
This is a retrospective from DatoCMS, a bootstrapped SaaS company, celebrating its 10th anniversary. The company reports achieving €6.5 million in revenue with a team of just 13 people, highlighting a remarkable 65% EBIT margin. The post reflects on sustainable, profitable growth as an alternative to the venture capital-funded "growth at all costs" model.
AI Democratizes Formal Verification: AI is lowering the barrier to entry for formal methods (Articles 1, 9). Tools like AI-assisted provers and educational frameworks are making it feasible to mathematically verify software correctness beyond niche, safety-critical systems. This matters because it could lead to more robust, secure, and reliable software at scale, shifting the industry paradigm from "test to find bugs" to "prove correctness from the start."
The Shift to On-Device & Efficient Inference: The focus on frameworks like Executorch (Article 5) underscores a major trend toward deploying smaller, optimized models directly on edge devices. This matters for user privacy, latency, cost, and functionality in disconnected environments. The implication is that the next wave of AI innovation will be in model compression, efficient architectures, and specialized hardware toolchains, not just raw model capability.
AI Integrates into Developer Toolchains: AI is becoming embedded in core developer tools, from proof assistants (Article 1) to CLI-based utilities (Articles 7, 8). This trend matters because it increases developer productivity and changes workflows. The takeaway is that successful AI/ML products will often be those that integrate seamlessly into existing workflows (like the terminal or IDE) rather than existing solely as standalone platforms.
The Rise of Declarative and Specification-Driven Development: Articles 1 (formal specs) and 7 (YAML-driven API tests) point to a trend where developers declare the what (the specification, the expected outcome) and let intelligent tools figure out the how (verification, execution). For AI/ML, this means a growing role for AI in interpreting high-level intent, generating tests, and ensuring systems meet their declared specifications automatically.
Sustainability and Profitability as a Counter-Narrative: The success of bootstrapped companies like DatoCMS (Article 10) offers a relevant counterpoint to the capital-intensive AI race. It matters because it highlights that sustainable economics and focused problem-solving remain viable. For AI/ML, the insight is that valuable applications can be built with discipline and deep domain expertise, not just massive compute and datasets, potentially guiding more entrepreneurs toward sustainable AI ventures.
Specialized Hardware and Software Co-design is Critical: The need for instant database clones for AI testing (Article 3) and on-device inference (Article 5) highlights that performance is increasingly dependent on deep software and hardware integration. This matters because pure algorithmic advances will be bottlenecked by I/O and compute efficiency. The trend pushes AI engineering towards expertise in systems programming, kernel features, and hardware-aware optimization.
AI Enables New Forms of Human-Computer Interaction (HCI): Concepts like the syntax-highlighting font (Article 6), while not directly AI, align with a trend of using smart systems to simplify complex digital experiences. AI can power similar "invisible" enhancements—anticipating user needs, simplifying interfaces, or personalizing workflows. The implication is that impactful AI/ML will often be felt in the removal of friction, not just in flashy new features.
Analysis generated by deepseek-reasoner