SShortSingh.

Programming

0
ProgrammingDEV Community ·

Cursor Launches Origin, a Git-Compatible Code Host Built for AI Agents

Cursor, the AI code editor now owned by SpaceX following a $60 billion acquisition that closed on August 14, 2026, launched a new code-hosting service called Origin on August 17, 2026. Origin is a Git-compatible platform built directly into the Cursor app, offering repositories, pull requests, code browsing, and two-way real-time sync with GitHub. Cursor's AI coding agents are embedded within every repository, allowing users to ask questions, make code changes, update pull requests, and push branches without leaving the app. The service integrates with Vercel, Depot, and Buildkite at launch, and is available exclusively to paid plan users across Pro, Teams, and Enterprise tiers. Origin's release notably coincided with an approximately eight-hour worldwide GitHub outage on the same day.

0
ProgrammingDEV Community ·

TCS, Persistent CEOs Confirm Shift to Outcome-Based Pricing Amid AI Productivity Gains

India's top IT firms are overhauling their decades-old billing model as AI reshapes how services are priced and delivered. In August 2026, TCS CEO K Krithivasan told Reuters that roughly 80% of the company's finance, HR, and business-services contracts are now priced on outcomes rather than hours worked. Persistent Systems CEO Sandeep Kalra separately confirmed that clients are demanding the same work for 25–30% less, while expecting faster delivery. TCS attributed the shift partly to AI-driven productivity gains of 10–15%, which it passes to clients in phases to protect its own margins. Cognizant also signalled the same direction, stating that clients now expect measurable business results rather than billable effort.

0
ProgrammingDEV Community ·

In 2026, AI Agent Infrastructure Is Solved — But Context Remains the Hard Problem

By mid-2026, building a functional AI agent has become straightforward, with platforms like Cloudflare Agents SDK and Vercel AI SDK absorbing what once required months of engineering work. However, agents continue to fail in production not due to model limitations, but because they lack organizational context — the decisions, discussions, and institutional knowledge that exist outside code and tickets. A July 2026 arXiv paper (arXiv:2607.14275) confirmed that context quality metrics such as grounding sufficiency and instruction consistency reliably predict agent reliability before deployment. Tools like MCP can grant agents access to data, but raw connector output often floods the context window without providing genuine understanding. The recommended fix is a dedicated context layer that retrieves, reconciles, ranks, and permission-scopes relevant knowledge before the agent begins reasoning.

0
ProgrammingDEV Community ·

AI-Heavy Firms Cut Jobs but Also Grew Headcount 10%, Study Finds

Major tech companies including Block, Oracle, and Meta collectively cut tens of thousands of jobs in 2026 while simultaneously increasing AI spending, contributing to over 121,000 tech layoffs tracked so far this year. However, a working paper analysing data from more than 21,000 US businesses found that heavy AI spenders actually grew their overall headcount by roughly 10% over two years, with entry-level hiring also rising. OpenAI's GDPval benchmark, released in September 2025, revealed that frontier AI models now compete directly with experienced professionals across 1,320 specialised tasks in 44 occupations. Researchers and analysts suggest the layoffs are concentrated in roles focused on routine task execution, while demand grows for workers who bring judgment and decision-making ownership. The emerging pattern indicates that AI is reshaping labour markets rather than simply shrinking them, with job security increasingly tied to the level at which a worker's value sits above what AI can automate.

0
ProgrammingDEV Community ·

Helios AI Model Delivers Real-Time Video at 19.5 FPS with 100x Cost Reduction

Helios is a 14-billion-parameter video generation model developed by researchers from Peking University, ByteDance, and Canva. The model runs on a single H100 GPU at 19.5 frames per second, cutting inference costs by approximately 100 times compared to previous approaches. It achieves this efficiency through architectural innovations including adversarial hierarchical distillation, which reduces sampling steps from 50 to just 3. Helios supports text-to-video, image-to-video, and video-to-video tasks through a unified input representation. The cost and latency improvements make interactive video applications — such as virtual try-ons, accessibility tools, and real-time content creation pipelines — economically viable for the first time.

0
ProgrammingDEV Community ·

TickDB MCP Lets Claude Access Live Market Data With Verified Timestamps

Developers can now connect Claude to real-time market data by configuring the TickDB Model Context Protocol (MCP) server, eliminating reliance on the AI's outdated training data. Without a live data connection, Claude reconstructs prices from memory with no timestamps, making it unreliable for financial research or trading workflows. MCP is an open standard that allows AI tools like Claude to call external APIs mid-conversation, similar to how they invoke a calculator or code interpreter. Once set up via Claude Desktop's config file using a TickDB API key, Claude can fetch live price snapshots, order book depth, historical candlestick data, and other market metrics with millisecond-precision timestamps. Users can independently verify the timestamps using a simple Python script, confirming exactly when the market data was recorded.

0
ProgrammingDEV Community ·

Node.js Guide Advocates Token-Scoped Event Trust for Healthtech IoT Dashboards

A technical guide published on DEV Community outlines a framework for building safer IoT control panels in healthtech environments using Node.js and real-time event discovery. The approach emphasizes that device presence should be treated as an observation rather than a permission, with each event type bound to a narrow token scope to prevent unauthorized actions. The author proposes classifying events into four families — presence, capability, intent, and outcome — to ensure dashboards do not conflate connectivity with authority. A Go-based event envelope structure is recommended, embedding authorization, sequencing, and trust fields directly alongside event names to make unsafe assumptions structurally difficult. The guide stresses that only server-signed observations should drive status displays, while browser-side callbacks must never be permitted to authorize device commands.

0
ProgrammingDEV Community ·

Rust MCP Tool Cuts Next.js Error Tokens by 99.7% While Blocking Credential Leaks

A developer has released an open-source Rust-based tool called Tokenectomy that compresses verbose Next.js and other framework error logs before they are sent to AI coding assistants. The tool reduces a typical 45,820-token error dump to just 118 tokens in under 0.2 milliseconds by stripping irrelevant node_modules stack frames and retaining only the actionable crash location. It also locally redacts secrets such as database passwords and API keys using a deterministic regex pipeline, preventing credentials from reaching cloud-based AI models. Built on the Model Context Protocol (MCP), it integrates with tools like Cursor and Claude Desktop via a pre-built binary distributed through npm, requiring no Rust installation. The tool runs entirely locally over stdio with a memory footprint of 3.45 MB and received a Grade A score from an independent audit by Glama.ai.

0
ProgrammingDEV Community ·

Three SQL Window Function Techniques to Solve the Gaps and Islands Problem

The 'gaps and islands' problem in SQL involves identifying consecutive streaks in data, such as uninterrupted user login days, and the breaks between them. Since SQL lacks a native streak-detection function, developers rely on combinations of window functions like ROW_NUMBER(), LAG(), and running SUM() to build solutions. The classic approach subtracts a sequential row number from each date, producing a shared computed value for rows within the same streak that shifts whenever a gap appears. A second, more readable method uses LAG() to flag rows that begin a new streak, then applies a cumulative SUM() to assign each streak a unique group ID. Both techniques, along with a third variation, ultimately produce the same output — the start date, end date, and length of each consecutive streak per user.

0
ProgrammingDEV Community ·

Developer Builds Token Forensics Tool After Claude API Costs Spike Unexpectedly

A developer repeatedly hit Claude Code quota limits and found that standard usage counters showed total consumption but not the underlying causes of waste. After analyzing 46 local session transcripts containing over 1,160 API calls and 276 million tokens, they identified around $10.47 in avoidable spending per roughly $187.80 total cost. Key waste patterns included cache misses re-sending large contexts, TTL downgrades mid-session, oversized tool outputs, and duplicate tokens from session splits. The findings led to the creation of an open-source CLI tool called quota-autopsy, which runs entirely locally with no external dependencies or telemetry. A free tier offers a cost overview, while a one-time $24 Pro upgrade unlocks detailed per-session breakdowns with suggested fixes.

0
ProgrammingDEV Community ·

How Combining AI Agent-Browser with Playwright Strengthens UI Testing

A development team integrated agent-browser into their UI evaluation workflow to test whether product goals are achievable, rather than verifying fixed click sequences. The setup uses AI coding agents like Codex or Claude to navigate the browser via agent-browser, which exposes an accessibility-based interface for inspecting and interacting with the UI. A key architectural decision was separating the agent's actions from result verification — the agent reports task completion, but Playwright independently checks the actual application state before a test is marked as passing. This dual-layer approach is described as exploratory QA, intended to complement rather than replace traditional end-to-end test suites. The team also outlined a parallel path using OpenAI's Computer Use for native desktop application evaluation, keeping the two workflows distinct.

0
ProgrammingDEV Community ·

Valve's Steam Frame VR Launches in Europe at €699: How It Compares to Rivals

Valve has officially priced its new Steam Frame VR headset at €699 in Europe, sparking debate over whether the cost is justified. The device features dual OLED displays at 2160×2160 pixels per eye, a Snapdragon XR2+ processor, 12 GB of RAM, and 256 GB of expandable storage, positioning it as a premium standalone headset. Compared to key rivals, it outperforms the Meta Quest 3 (€499) and Pico 4 (€449) on specs but costs less than the HTC Vive XR Elite (€799). When accessories are factored in, the total estimated cost rises to around €824, though summer 2026 discounts of up to 20% from major European retailers could bring that figure closer to €660. Analysts suggest the headset is best suited for users already invested in the Steam ecosystem, while budget-conscious buyers may find better value in the Quest 3 or Pico 4.

0
ProgrammingDEV Community ·

How to Test Whether a Free API Tier Can Handle Your AI Agent's Real Workload

Engineering teams often migrate agent jobs to free-tier endpoints without measuring whether those endpoints can sustain production-level loads over time. Free tiers typically degrade silently through increased latency, retry amplification, and queue buildup rather than outright failures. A structured load-testing approach using four key metrics — p50/p95 latency, error rate, cold-start penalty, and batch drain time — can reveal fitness problems before they affect operations. A sample Python probe script is provided as a template harness to measure tail latency and error rates against any JSON-based model endpoint. The article was produced as part of outreach for MonkeyCode, an open-source coding agent that offers free model access and a free server option.

0
ProgrammingDEV Community ·

How to Safely Migrate E-Commerce DNS Zones Using Enumerate-Diff-Apply-Verify

A structured DNS migration approach for e-commerce platforms recommends four ordered steps: enumerate live records, diff them against the intended configuration, upsert only the differences, and verify outcomes before switching nameservers. Skipping the diff phase risks silently deleting records, turning a missed entry into a live incident rather than a reviewable change. The method distinguishes between customer-owned zones, which require an import-and-approval boundary, and platform-owned zones that can follow standardized templates. Raw DNS responses should be saved before any normalization, as this snapshot serves as rollback material if the migration needs to be reversed. The author recommends a lightweight Python control plane over inheriting a legacy Node.js registrar SDK, keeping the migration logic transparent and vendor-agnostic.

0
ProgrammingDEV Community ·

Study Finds N-gram Count Tables Lack Memory Locality, Defying Earlier Assumption

A software developer re-examined a core assumption in a series of language model experiments: that count tables used for next-token prediction would behave like mixture-of-experts models, keeping only a small 'hot set' of data resident in memory. To test this, the researcher rebuilt the tables as memory-mappable, pageable arrays and verified they produced identical accuracy results to the original in-memory versions. Measurements across eight real code files showed that working set memory kept growing throughout roughly 2,700 completion positions per file, with seven of eight files still expanding at the end of the session. A bounded cache test confirmed the problem, with hit rates plateauing at only 74.4% even when 16 MB was held resident, and dropping to 47.3% at 1 MB. The findings show that while code tokens repeat frequently, the count table is indexed by context rather than raw tokens, meaning locality at the token level does not translate into locality at the lookup level.

0
ProgrammingDEV Community ·

Developer Replaced LLM-Driven Interview Flow with a Finite State Machine for Reliability

A developer building an AI-powered technical interviewer found that relying solely on a large language model to manage interview flow caused consistent failures, including premature scoring, repeated questions, and accidental answer giveaways. The core problem was structural: LLMs lack reliable state-tracking across conversation turns, making them unsuitable as sole controllers of multi-stage workflows. To fix this, the developer implemented a finite state machine that owns all phase transitions, such as moving from coding to evaluation only after code is actually submitted and run in a sandbox. The LLM was retained for content generation within each state — writing questions, hints, and scorecards — but stripped of any authority to decide what happens next. The result was a more reliable and, counterintuitively, more naturally conversational product, since the LLM could be expressive within tightly controlled boundaries.

0
ProgrammingDEV Community ·

Rails structure.sql Explained: The Auto-Generated DB Snapshot That Speeds Setup

In Ruby on Rails projects, the file db/structure.sql serves as a complete SQL snapshot of the current database schema, capturing all table definitions and applied migrations in one place. Rails automatically generates and updates this file each time the db:migrate command is run, meaning developers should not edit it manually. Its primary advantage over replaying all migration files is speed: rather than executing hundreds of incremental change scripts one by one, a new or test database can be built instantly from the single finished blueprint. Migration files remain essential for safely applying incremental changes to live production databases that already hold data. Projects requiring database-specific features like triggers — which cannot be expressed in Rails' default Ruby-based schema.rb — typically opt for structure.sql instead.

0
ProgrammingDEV Community ·

How to Build Deterministic AI Agent Loops Using Finite State Machines in TypeScript

Software engineers building production AI agents frequently encounter failures such as infinite loops, hallucinated arguments, and costly unvalidated state mutations that work in demos but break in real deployments. A technical guide published on DEV Community proposes replacing naive LLM agent loops with a Finite State Machine architecture featuring isolated verification gates and event-sourced replay. The approach introduces a four-phase execution model called Observe, Propose, Verify, and Commit, which decouples LLM planning from deterministic runtime validation before any tool is executed. Each phase is enforced through typed contracts using Zod schemas and TypeScript, ensuring that invalid state transitions are caught and logged before they cause side effects. The architecture aims to make AI agent runtimes production-ready by preventing context pollution, non-idempotent mutations, and unbounded retry loops that drain API credits.

0
ProgrammingDEV Community ·

How Karpenter Fixes EKS Node Scaling Without Predefined Node Groups

Traditional EKS autoscaling relies on Cluster Autoscaler paired with Auto Scaling Groups, requiring teams to pre-define node groups tied to specific instance types — a system that often leads to wasted capacity and complex configurations. Because Cluster Autoscaler can only simulate scaling against existing node groups, mismatches between pod resource requests and instance sizes frequently result in stranded CPU or memory. Managing multiple node groups across instance families, capacity types, and availability zones can balloon into dozens of configurations that quickly become outdated. Karpenter, an open-source autoscaler for EKS, takes a fundamentally different approach by evaluating pending pods at scheduling time and dynamically selecting the most suitable instance to launch, rather than scaling a predefined group. It also batches unschedulable pods within a short window before acting, enabling more efficient and consolidated provisioning decisions.

← NewerPage 1324 of 3309Older →