0
ProgrammingHacker News ·
Economic Policy for AGI
Article URL: https://institute.deepmind.com/essays/economic-policy-for-agi/ Comments URL: https://news.ycombinator.com/item?id=49743685 Points: 3 # Comments: 0
Programming
0
ProgrammingDEV Community ·
How to Build a Reliable Performance Test for Beauty AR Sessions in TypeScript
A Beauty AR proof of concept may perform flawlessly in short demos but fail under real-world sustained usage, making rigorous testing essential before shipping. Effective testing requires running representative effect combinations over extended sessions to detect gradual performance degradation, not just measuring average frame rates. Developers should define test cases based on actual product configurations — such as baseline, common, expensive, and recovery profiles — rather than synthetic stress tests that don't reflect user behavior. Device tiers should be maintained using real-device benchmarks rather than user-agent strings, with an unknown tier defaulting to conservative settings. A TypeScript-based testing framework using a platform-agnostic adapter is recommended to keep SDK-specific calls separate and portable across target platforms.
0
ProgrammingDEV Community ·
BS9 Is an Open-Source Self-Healing Process Manager Built for Bun
A developer has released BS9 (Bun Sentinel 9), an open-source process supervisor and high-availability clustering manager designed specifically for the Bun JavaScript runtime. The tool addresses common production challenges such as zero-downtime reloads, fault tolerance, and cross-platform daemon management that existing tools like PM2 handle poorly for Bun. BS9 uses a Replace-First reload strategy that spins up a new worker, confirms its readiness, drains in-flight requests, and only then shuts down the old process to avoid dropped traffic. It also includes a built-in State Hub with write-ahead logging and compare-and-swap primitives, eliminating the need for an external Redis cluster. A built-in chaos-testing command, verify-ha, lets developers stress-test their applications with simulated crashes and rolling reloads to measure latency and availability.
0
ProgrammingDEV Community ·
Kubernetes Sits Above VMs, Not in Place of Them, Experts Clarify
A common misconception in enterprise IT holds that Kubernetes replaced virtual machines, but the two technologies operate at different architectural layers and are designed to coexist. Kubernetes functions as an application-level control plane, handling workload scheduling, reconciliation, and service discovery, while virtualization continues to manage compute abstraction underneath. The real shift was not a replacement of VMs but a consolidation of fragmented application management responsibilities into a single higher-level system. Before Kubernetes became standard, tasks like application placement and lifecycle management were split across VM tooling, platform scripts, and manual operator decisions. Misunderstanding this distinction can lead organizations to make flawed platform transition decisions, such as focusing on replacing a hypervisor while overlooking the broader operating model that depends on it.
0
ProgrammingDEV Community ·
Why Frameworks Are More Than Code: They Encode Decades of Hard-Won Security Lessons
A developer reflecting on 1980s bulletin board systems recalls storing user passwords in plaintext on floppy disks, a practice that exposed serious security flaws over time. Each attempted fix — from substitution ciphers to homegrown transformations — revealed new vulnerabilities, gradually building decades of engineering knowledge around something as seemingly simple as password storage. Modern authentication now encompasses hashing, salts, rate limiting, CSRF protection, session management, and more. The author argues that mature software frameworks do far more than save developers from writing code — they encode institutional memory, embedding hard-learned decisions, discovered edge cases, and patched vulnerabilities. A first-time developer implementing authentication may not know the right questions to ask, but a well-built framework often already has the answers.
0
ProgrammingDEV Community ·
How Tapflow Solved Async Clipboard Sync Between Browser and Mobile Simulators
Tapflow, a browser-based device viewer, faced a two-part clipboard challenge: device clipboards operate asynchronously, and browsers enforce permission checks before a response from the device arrives. For paste, the browser captures the paste event and relays the text to a device agent, which writes it to the simulator's clipboard and triggers a platform-specific paste input — Cmd+V on iOS and KEYCODE_PASTE on Android. For copy, a fixed delay proved unreliable, so Tapflow instead writes a random sentinel value to the device clipboard, waits until it is confirmed visible, then polls for a changed value to confirm the app's copy command actually landed. If the copy fails, the original clipboard content is restored and verified before the device is unblocked for further clipboard operations. An additional browser constraint — Safari rejecting clipboard writes after the original user gesture expires — further shaped the final implementation approach.
0
ProgrammingDEV Community ·
How Valid Markdown Can Hide Costly Arithmetic Errors in Business Documents
A narrative from DEV Community illustrates how Markdown-based invoices can appear visually correct while containing serious calculation errors. In the example, a project manager updates a line-item quantity from 12 to 20 hours but leaves all derived figures — net amount, VAT, gross total, and payment schedule — unchanged. Both Git and the Markdown renderer flag no issues, since neither tool understands that figures should satisfy formulas like Net = Qty × Rate. A reviewer approves the diff without recalculating, because no individual number looks obviously wrong. The piece argues that Markdown's lack of constraint enforcement makes it a structurally fragile format for financial documents, especially in fast-growing small teams where informal processes outlast their original low-stakes context.
0
ProgrammingHacker News ·
Insufficient source material to produce a verified summary
The source article contains no substantive text beyond a job listing URL and a Hacker News thread link. No factual details about the role, company background, hiring scope, or context were provided in the article body. Generating a summary would require fabricating information, which falls outside editorial guidelines. Please provide a fuller article text for accurate reporting.
0
ProgrammingDEV Community ·
Claude Code Headless Mode Enables Autonomous CI/CD Execution via -p Flag
Claude Code's headless mode allows developers to run AI-assisted coding tasks autonomously without interactive terminal sessions, using the -p flag to pass prompts directly as command-line arguments. This approach is designed for production environments where repeatable, scheduled, or parallel execution is required across multiple repositories or microservices. Key features include output format control (including structured JSON), session persistence options, and the ability to chain results into downstream build steps. Engineering teams can coordinate multiple parallel headless agents using TypeScript and Node's child_process module, with each agent handling errors independently. The shift from interactive to headless execution moves operational responsibility from manual supervision to robust error handling and retry logic within CI/CD pipelines.
0
ProgrammingHacker News ·
I didn't sign the Fields medallists' letter
Article URL: https://terrytao.wordpress.com/2026/09/17/why-i-didnt-sign-the-fields-medallists-letter/ Comments URL: https://news.ycombinator.com/item?id=49743534 Points: 5 # Comments: 0
0
ProgrammingDEV Community ·
Treat Gemini Structured Output as a Versioned API to Prevent Agent Breakage
When applications rely on structured JSON output from Gemini, that output effectively becomes an API contract, meaning field changes or new enum values can silently break downstream consumers. Developers are advised to embed schema version identifiers directly in persisted output values rather than relying solely on deployment configuration. Google's Gemini SDK supports JSON Schema alongside Zod validation, enabling stricter runtime checks at the application boundary. Structured generation handles syntax and shape validation, but semantic and policy-level checks — such as verifying referenced evidence IDs exist or confirming an action is permitted — remain the application's responsibility. Different validation failures require different handling strategies, and policy violations in particular should be blocked rather than re-prompted until the model produces a compliant answer.
0
ProgrammingHacker News ·
Detail.dev Blog Explores the Concept of Self-Driving Codebases
A blog post published on detail.dev examines the emerging idea of self-driving codebases, where software systems could autonomously manage and evolve their own code. The article explores how AI and automation tools might reduce the need for manual developer intervention in routine codebase maintenance. The piece appears to be part of a broader conversation in the developer community about increasing AI involvement in software engineering workflows. The post received modest engagement on Hacker News, accumulating 5 points and one comment since its submission.
0
ProgrammingDEV Community ·
WSL2 and Docker silently consume Windows C drive space and never auto-release it
Windows developers using WSL2 or Docker Desktop often face unexplained disk space loss because both tools store data in virtual disk files (ext4.vhdx or docker_data.vhdx) that grow automatically but never shrink on their own. Deleting containers, running docker system prune, or removing WSL projects does not reclaim space on the Windows C drive, as the freed gigabytes remain trapped inside the virtual disk file. A related Microsoft WSL GitHub issue has been open since 2019 and has accumulated over a thousand reactions, indicating the problem is widespread among developers. Users can manually reclaim space by fully shutting down WSL and Docker, then compressing the .vhdx file using either the Optimize-VHD PowerShell command (Windows Pro/Enterprise) or diskpart (Windows Home). Beyond virtual disks, dev machines also quietly accumulate space from forgotten node_modules folders, build caches, and IDE data that standard Windows storage tools do not clearly flag for cleanup.
0
ProgrammingHacker News ·
Infinite-Parameter LLMs: Generating and Adapting Weights from Live Data
Article URL: https://arxiv.org/abs/2609.18842 Comments URL: https://news.ycombinator.com/item?id=49743483 Points: 6 # Comments: 0
0
ProgrammingHacker News ·
AutoBot: Open-Source Voice Controller for Long-Running AI Agent Tasks
A developer has released AutoBot, an MIT-licensed open-source tool designed to manage long-horizon AI workstreams through native voice commands. The project allows users to initiate tasks, monitor progress, and steer AI agents hands-free, while a local ledger tracks incomplete outputs and completion criteria. AutoBot stores data on encrypted local disk to support deep personalization while maintaining strict privacy, and uses a nightly learning consolidation process alongside a heartbeat system to persist execution. The tool was benchmarked on OSWorld, scoring 32.41%, and on AssistantBench, scoring 50.70%, without requiring virtual machines or significant setup overhead. The creator built it as a personal project to replicate a Jarvis-like assistant and has shared the code on GitHub, inviting others working on similar tools to collaborate.
0
ProgrammingHacker News ·
Developer Launches Marketplace for Paid AI Agent Skills, Curated by Humans
A developer known as Nick (@skeptrune on X) has launched a platform described as a 'Craigslist for agent skills,' designed to let users buy specialized skills to enhance AI agents. The idea was inspired by Nick's own experience using AI for tasks outside his expertise, such as contract redlining, AI video creation, and website design. The platform is human-curated, distinguishing it from fully automated skill marketplaces. Nick shared the project on Hacker News, targeting an engineering and AI-coding-agent audience he believes faces similar challenges. The post received modest early engagement with 6 points and 3 comments at the time of publication.
0
ProgrammingHacker News ·
Aclif Open-Source Tool Brings Unified CLI Grammar to SaaS Platforms
A developer has released Aclif, an open-source agent CLI framework designed to standardize command-line interactions across multiple SaaS providers. The tool uses a single grammar and canonical naming conventions, aiming to reduce friction when working with different platforms. Default provider schemas are embedded directly in the binary, allowing basic discovery without requiring credentials. Users can optionally supply organizational credentials to fetch instance-specific customizations. The project is publicly available on GitHub for developers to explore and contribute.
0
ProgrammingDEV Community ·
Developer delivers colleague's tech talk at Startup Summit, learns knowledge transfer in practice
In late August 2025, a developer at Dati Labs delivered a talk titled 'Price per token is not price' at Startup Summit 2026 in Florianópolis, originally created by colleague Bruno Vilardi Bueno. Rather than simply reusing the existing slides, the presenter studied the material thoroughly, re-ran live demos on their own account, and internalized the content before taking it to the audience. The experience highlighted a broader organizational risk: when only one person can deliver a piece of content, it becomes a bottleneck regardless of its quality. The presenter also discovered new data not in the original material — the same sentence produced 25% more tokens in Anthropic's model family 5 than in family 4.6, underscoring how quickly AI pricing details shift. The exercise illustrated what researchers Nonaka and Takeuchi describe as the knowledge spiral, where externalized knowledge only becomes an asset once someone else fully internalizes it.
0
ProgrammingDEV Community ·
AI Hyperscalers' Debt Binge Blamed for Pushing Long-Term Treasury Yields Higher
Federal Reserve Chairman Kevin Warsh identified the surging debt issuance by AI hyperscalers as a contributing factor to elevated long-term Treasury yields, with the 10-year note breaking above 5 percent for the first time since 2007. Corporate bond issuance by the five largest American hyperscalers jumped from an average of $28 billion annually between 2020 and 2024 to $121 billion in 2025, according to Bank of America Securities. Morgan Stanley estimates global AI-related debt reached $236 billion by late spring 2025 and could approach $570 billion through 2026. The borrowing surge is driven by capital expenditure on data centers and AI accelerator hardware that now consumes nearly all operating cash flow for several cloud providers. Analysts warn of a reflexive dynamic in which the AI sector's own debt issuance raises the benchmark discount rate, thereby increasing the financial hurdle its investments must clear to be justified.
0
ProgrammingDEV Community ·
Developer Builds FinderBack to Add Navigation Shortcuts to macOS Finder Menu
A developer has released FinderBack, a small open-source utility for macOS that adds Back and Forward navigation options to Finder's right-click context menu. The tool was created to eliminate the repetitive need to move the cursor up to the toolbar for basic navigation. Users can right-click anywhere within Finder to access the navigation commands directly. FinderBack is available both as source code on GitHub and as a ready-to-install build via the project's website.