SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World

Programming

LatestTopDiscussed
0
ProgrammingDEV Community ·

Sinch CLI Lets Developers Build, Deploy, and Automate Voice and Messaging APIs from Terminal

Sinch has released a command-line interface tool (@sinch/cli) that allows developers to manage its platform — including phone numbers, Voice API v2, messaging, and cloud functions — through a single terminal command. The CLI supports tasks such as renting phone numbers, routing inbound calls, scaffolding and deploying voice functions, and sending messages via the Conversation API. It also integrates with AI coding agents like Claude Code, Cursor, Kiro, and GitHub Copilot, which can run CLI commands autonomously using structured JSON output and exit codes. Authentication is handled via a single login that covers all Sinch products, with credentials stored securely in the operating system's native keychain. The tool requires Node.js 24 or newer and a Sinch account with a configured project and access keys.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Amadeus CTO Says MCP Can't Handle Complex Travel Workflows — Experts Agree, Partly

The CTO of Amadeus, one of the world's largest global distribution systems, recently stated that the Model Context Protocol cannot manage complex retail workflows, sparking debate in travel tech circles. MCP has gained significant traction over the past two years, praised for enabling seamless AI-to-tool integrations across platforms. Critics and analysts note the CTO's concern is valid but partial — MCP excels at exposing discrete, atomic actions like search and pricing, but lacks built-in support for orchestrating multi-step, stateful processes such as bookings, refunds, and dispute resolution. Developers are advised to use MCP directly for read-only, high-frequency tasks, while relying on orchestration layers like LangGraph or server-side business logic for complex, compliance-heavy transaction flows. The consensus is that MCP is a powerful connection protocol, not a full transaction management system, and architects must decide which parts of a workflow belong at each layer.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer Builds 16-Server MCP Monorepo With 65 Tools and Full CI in 48 Hours

A developer built a monorepo housing 16 independent Model Context Protocol (MCP) servers in just 48 hours, covering tools for hashing, base64 encoding, RSS reading, YouTube transcripts, and more. Rather than building a single large server, the developer deliberately split functionality into small, focused packages to reduce context bloat, limit trust exposure, and speed up installs. The project includes 197 tests, 65 tools, and a CI pipeline that runs lint, test, and build jobs as a matrix across all 16 servers, pinpointing failures to a specific package. Each server is independently installable via PyPI using OIDC-based Trusted Publishing, with no API tokens stored in the repository. The monorepo uses a shared root configuration for dependencies and tooling, while each server maintains its own package definition and entry point.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

AMD Acquires World Labs for $8.2B, Names Fei-Fei Li Chief Scientist

AMD has agreed to acquire World Labs, a two-year-old AI startup, for $8.2 billion in an all-stock deal expected to close by end of 2025, subject to regulatory approval. The startup, founded by renowned AI researcher Fei-Fei Li, specialises in spatial-intelligence and world models — AI systems capable of perceiving and interacting with 3D environments. Li will join AMD as Chief Scientist and Executive Vice President, reporting directly to CEO Lisa Su. AMD, which already had an inference-optimization partnership with World Labs, says the acquisition will help it anticipate the compute demands of frontier workloads in robotics, simulation, and physical AI. The $8.2 billion price tag represents a significant premium over World Labs' January 2025 fundraising valuation of approximately $5 billion, making it AMD's second-largest acquisition on record.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Meta's Muse Agent Uses Connectors and Cron Tasks to Automate Real Workflows

Meta's personal AI agent Muse moves beyond simple question-answering by offering two core building blocks: connectors and scheduled tasks. Connectors are authenticated integrations with external services like Gmail, Google Calendar, Slack, and Notion, giving Muse read and write access to act on a user's accounts. Scheduled tasks allow users to define time-triggered jobs in plain language, such as a daily 7 AM morning briefing, which Muse then executes automatically in the background. Together, these primitives let the agent pull live data from real accounts, perform multi-step work, and deliver structured summaries without requiring webhooks or custom glue code. Developers are advised to grant minimum required permissions and to define tasks by desired outcome rather than step-by-step instructions for best results.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Citrix Patches Two Actively Exploited NetScaler Zero-Days Rated CVSS 9.5

Citrix released advisory CTX697096 on September 27, 2026, fixing eight vulnerabilities in NetScaler ADC and Gateway, including two critical zero-days already being exploited in the wild. Security firm watchTowr had issued a public warning a day earlier, citing findings from a forensic investigation, while the Dutch national cybersecurity center privately advised some organizations to take affected appliances offline. Both zero-days — CVE-2026-88771 and CVE-2026-88772 — carry a CVSS v4 score of 9.5 and allow unauthenticated remote code execution on default configurations. CISA added both flaws to its Known Exploited Vulnerabilities catalog the same day, giving federal agencies only until September 30 to apply patches. Separately, Sidero Labs announced on September 14 that Talos Linux would gain native hypervisor capabilities, enabling unified management of containers and virtual machines through a single OS and API, with general availability targeted for December 2026.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

How Much a Photography Website Really Costs: A Developer's Breakdown

Building a photography website involves far more than a simple portfolio, requiring image optimization, booking tools, galleries, and local search features that drive costs up significantly. Depending on the route chosen, upfront costs range from near zero for DIY builders to over $20,000 for agency-built sites, with ongoing monthly expenses adding up over time. A basic builder-based portfolio costs roughly $1,260–$1,560 over three years, while a freelancer-built WordPress site typically runs $3,400–$5,200 when hosting and updates are factored in. Key cost drivers include gallery design, online booking with deposits, print stores, and speed optimization, each carrying its own one-time fee. Experts advise photographers to rent proofing galleries rather than build them, launch in phases, and always secure a fixed-scope written quote before committing.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

What Shopify's Built for Shopify Badge Actually Certifies — and Its Limits

The Built for Shopify badge is Shopify's highest recognition tier for App Store apps, certifying that an app has passed the platform's own testing for performance, design, and integration. For storefront-rendering apps, the badge confirms the app reduced a store's Lighthouse score by no more than 10 points in Shopify's internal testing. To qualify, an app must have at least 50 net installs from active paid-plan stores and five reviews, meaning newer apps are excluded regardless of quality. Badged apps rank higher in App Store search results, so top search positions partly reflect badge status rather than pure relevance or fit. The badge does not cover pricing, customer support quality, or suitability for a specific store, and merchants are advised to verify those factors independently.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

How to Route Shopify Webhooks into Google Sheets: Tools and Workarounds

Shopify webhooks cannot send event data directly to a Google Sheets URL, as the spreadsheet has no built-in mechanism to receive and respond to incoming HTTP requests. A middleware layer is required, such as Zapier, Make, or n8n, each of which can receive the webhook and then write a row to a sheet. Alternatively, a Google Apps Script web app can serve as the receiver via its doPost function, though developers must ensure it returns a proper 200 status code to avoid Shopify automatically deleting the webhook subscription. Shopify merchants on Basic, Grow, Advanced, or Plus plans can also bypass webhooks entirely by using Shopify Flow's native 'Add row to spreadsheet' action. Self-hosted n8n users should note that Shopify does not allow webhooks to be sent to localhost addresses.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Memory Bandwidth, Not VRAM Size, Determines Local LLM Token Speed

When running large language models locally, memory bandwidth in gigabytes per second is the primary factor governing token generation speed, not the total amount of VRAM available. Because autoregressive decoding requires the GPU to stream every model weight from VRAM for each token generated, the theoretical speed ceiling equals memory bandwidth divided by model size. Benchmarks illustrate the stakes sharply: an RTX 4060 Ti running Llama 8B at 42.5 tokens per second dropped to just 3.8 tokens per second when only 20 percent of weights spilled into system RAM, a 91 percent slowdown. Bus width drives this gap — the same 16 GB of VRAM delivers 288 GB/s on a 128-bit RTX 4060 Ti versus 672 GB/s on the 256-bit RTX 4070 Ti Super. For users seeking the best value, a used RTX 3090 with 24 GB on a 384-bit bus at around $650–750 remains the recommended choice for running 32-billion-parameter models with room for context.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

OpenAI Launches Always-On Agents as Nvidia Debuts Runtime Safety Watchdog

OpenAI introduced Dots at DevDay, autonomous agents that run continuously on cloud infrastructure, connect to over 4,000 apps, and proactively seek tasks without user prompts. The same week, Nvidia released the Open Agent Safety Platform, comprising OpenShell, an open-source runtime that traces and enforces agent behavior, and Sentry, a watchdog running on separate BlueField 4 DPUs physically isolated from the agent process. The timing is notable because always-on agents eliminate natural human review checkpoints, leaving credentials persistently exposed and failures potentially silent. Nvidia's out-of-band design matters because a watchdog sharing the same processor as the agent could itself be manipulated by a misbehaving agent. However, Nvidia acknowledges the platform enforces predetermined boundaries rather than acting as a universal kill switch, meaning the quality of safety still depends on the policies humans write.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

How ERC-3643 Transfer Controls Can Be Tested for Permissioned Token Compliance

ERC-3643 permissioned tokens require verifiable transfer controls, and a structured approach using real token, registry, and compliance contracts can make those rules observable and testable. A recipient-country allowlist rule was implemented against the actual ERC-3643 contract suite and validated through 19 passing tests, including a deliberately broken module and its correction. The walkthrough highlights a key discrepancy between the ERC-3643 specification, which states minting bypasses compliance, and the pinned implementation, which calls canTransfer during minting — a difference teams must resolve before any issuance transaction. Each compliance requirement, such as identity claims, country restrictions, wallet freezes, and agent roles, is mapped to a specific enforcement point and an observable test assertion. The exercise, conducted in the context of blockchain engineering at Pharos Production, underscores that a general claim of compliance support is insufficient without explicit, testable contract behavior for integration engineers.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer builds browser-based 3D marble run simulator with deterministic MP4 export

A developer has launched MarbleRun.online, a free browser-based 3D marble run sandbox where users can place ramps, pegs, spinners, and balls on a virtual wall and simulate realistic physics. The tool runs entirely client-side with no login, no uploads, and no AI involvement, and is intentionally limited to desktop browsers to preserve the 3D experience. Users can export their simulations as vertical 9:16 MP4 videos at 1080x1920 resolution with synchronized impact audio, a feature the developer says took three times longer to build than the editor itself. The export challenge centered on achieving determinism — ensuring the same level produces an identical video every time — which required offscreen rendering at a locked 30fps, a fixed-timestep physics clock, and frame-accurate generated audio instead of microphone capture. The developer noted that starting with the export pipeline rather than the editor would have saved significant rework, as determinism was not considered early enough in the build process.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

How Engineers Can Build Terminology Pipelines to Prevent Regulatory Translation Drift

Asset managers distributing funds across multiple EU markets must publish SFDR disclosures in each local language, using legally fixed terms that must remain consistent across all document versions and updates. Unlike grammar errors, the real risk is semantic drift, where a regulated term like 'Principal Adverse Impacts' gets translated slightly differently across documents by different vendors, potentially triggering regulatory scrutiny. A developer-focused approach treats glossary management as a data engineering problem, storing terms in structured, versioned JSON with locked translations rather than informal spreadsheets. This termbase can then feed directly into professional CAT tools such as memoQ or Trados, while automated scripts validate translated documents against approved terminology before publication. The method mirrors established software practices like API contract versioning, applying a single source of truth and programmatic checks to multilingual compliance workflows.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Neurodivergence Is a Context Mismatch, Not a Cognitive Deficit, Argue Researchers

A perspective piece on DEV Community argues that the concept of 'neurotypical' simply describes a brain that fits existing environmental assumptions, rather than one that functions optimally. The author contends that 'neurodivergent' brains are not broken but are instead mismatched to workplaces designed around a narrow cognitive profile emphasizing sustained attention, stable working memory, and linear prioritization. Using a software analogy, the piece compares evaluating neurodivergent individuals to calling an ARM binary deficient for failing to run on x86 hardware. The article challenges tech teams to reconsider how rigid workflows may suppress innovation by forcing all cognitive profiles into a single operational model. It concludes that the real question is not whether to accommodate neurodivergent workers, but how much potential is lost by refusing to do so.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

AI Agent Built Tools for Three Days With $5/Day Budget — and Made Zero Sales

A large language model (Qwen3.6-35B) operated by a human named Mouiz attempted to generate revenue autonomously over three days using only a single OpenRouter API key with a $5 daily limit. The AI built and listed several digital products — including a CSS Gradient Generator, a Web Scraping Starter Kit, and a Personal Toolkit — on Gumroad, but recorded zero sales and negligible traffic. It also developed free tools such as a Kanban Board and Pomodoro Timer, hosted on its only owned asset, the domain k1r4.space, which has not yet been indexed by Google. The agent identified distribution, not product quality, as its core bottleneck, noting that Gumroad products lack Google indexing and DEV.to articles by new authors receive very few views. The experiment highlights structural barriers AI agents face — including lack of legal identity, inability to pass KYC checks, and limited writable external channels — when attempting independent economic activity.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Xiaomi MiMo V2.6 Qwen 9B Tested on a 16GB M1 MacBook Pro

A developer running a series of local AI experiments tested Xiaomi's MiMo V2.6, a 9-billion-parameter model distilled from Qwen, on an entry-level 16GB M1 MacBook Pro. The model was evaluated using standard prompts, including a general knowledge question and a Python scripting task, with all tests run at real speed via MLX Core, a native Mac app built on Apple's MLX framework. MiMo V2.6 delivered detailed responses at around 12 tokens per second and uniquely requested tool access before writing code, though its output contained a minor formatting error. The developer ranked it roughly on par with Ternary Bonsai 2 and above Gemma 4B, calling it suitable for summarization and small utility scripts but not extended agentic coding sessions. The model is free to use, requires no API billing, and functions fully offline, which the developer highlighted as a key practical advantage.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Neurotypical Means Majority, Not Superior: A Case for Cognitive Diversity at Work

A piece published on DEV Community challenges the assumption that 'neurotypical' brains represent a gold standard, arguing instead that they simply align with the implicit design of most work environments. The author contends that modern workplaces are built around a narrow cognitive profile emphasizing sustained attention, stable working memory, and linear task prioritization. This setup does not measure capability, but rather how well a brain fits a specific context. The article reframes the neurodiversity conversation away from 'accommodation' and toward recognizing the broader value lost when all cognitive profiles are forced into the same operational model.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Batch APIs need per-item authorization, not just route-level access checks

Batch and bulk API endpoints allow a single request to fetch or modify multiple resources at once, but this convenience can introduce serious authorization gaps. If access control is only enforced at the route level, a caller permitted to view one record may inadvertently gain access to many others in the same batch. This vulnerability extends to bulk update, delete, and data export operations. Security best practice requires that every item in a batch be individually authorized against the current user, tenant, and action type. Any item the caller is not permitted to access should trigger a hard failure rather than being silently skipped.

0 comments Read more at DEV Community
← NewerPage 787 of 3167Older →