SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World

Programming

LatestTopDiscussed
0
ProgrammingDEV Community ·

Microsoft completes pre-training of 2.8-trillion-parameter MAI-Base-2 model, plans partner licensing

Microsoft AI announced on 15 October 2026 that pre-training of MAI-Base-2, a mixture-of-experts foundation model with 2.8 trillion total parameters and 80 billion active per token, is finished. The run used 16,000 Nvidia GB200 GPUs over 38 days in August and September 2026. Microsoft says its training recipe is more FLOPs-efficient than DeepSeek-V4-Pro-Base and its earlier MAI-Base-1, though those comparisons rest on scaling-law fits that include the model's own score. The model is now moving into post-training, and Microsoft plans to license its weights to trusted partners through Foundry, with no open-weights release, price, or date announced.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Building a Context-Aware Customer Support Agent on MongoDB Atlas

A tutorial by Samuel Molling describes how to build a customer support agent using MongoDB Atlas. The core is a retrieval-augmented generation pipeline that combines vector and lexical search, merges the ranked results with Reciprocal Rank Fusion, and reranks candidates with a cross-encoder before an LLM writes the answer. The author argues that real support conversations are context-dependent, so follow-up questions like "how much is that?" cannot be searched on their own and must be rewritten using conversation history. Answers may come from documentation, resolved past tickets, or customer account data, and some requests must be escalated to a human. The article therefore adds query rewriting, conversation memory, multiple data sources, customer entitlements, operational tools, confidence scoring, and escalation logic around the retrieval layer.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Using InstancedMesh in Three.js to cut draw calls for repeated parts

Building repeated geometry such as pergola louvres or fence slats with one Mesh per part can drag frame rates down on mid-range phones, since each object costs a separate draw call per frame. The author reports a configurator running at 12 FPS that reached a steady 60 FPS after switching the 40 louvres to a single InstancedMesh. The approach requires calling updateMatrix before setMatrixAt and setting instanceMatrix.needsUpdate after changing transforms, with per-instance colours handled through setColorAt and instanceColor.needsUpdate. The author suggests the technique pays off for roughly 100 to 200 or more repeated objects, or for objects updated every frame, while plain meshes are simpler below that threshold.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Four types of health AI each demand a different kind of engineering

A senior engineer building a health AI argues that the term covers four distinct products, each failing in its own way. Ambient scribes that write visit notes can invent statements, so the author would require clinician review before anything enters the chart. Clinician Q&A tools with citations can link to sources that do not actually support the claim, calling for automated checks of each claim against its cited passage. Patient-facing copilots that flag too much risk being ignored, so their flags need measured precision and recall. The author's own free health-information assistant, Tabibu, is designed around preventing false reassurance and dosing advice, and its limits are published openly. The piece also cites a 2026 Mount Sinai study showing a short safety reminder reduced potentially harmful clinical choices from 16.6% to 10.1%, which the author presents as progress but not a complete fix.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Unverified AI-generated code accumulates technical debt faster, experts warn

An opinion piece argues that teams adopting AI coding tools without changing their verification practices have not gained speed but have instead accumulated hidden risk. As AI has shrunk the time needed to write code, the bottleneck has shifted to confirming that the code actually works. The author cites Google's DORA 2024 report, which linked greater AI adoption to gains in code quality and review speed but to lower delivery throughput and stability. A Stack Overflow 2025 survey found that developers most often complained about AI output that is nearly correct but flawed, and that more of them distrust it than trust it. The piece also points to a 2025 METR trial in which experienced developers were slower with AI tools while believing they were faster, and it notes that METR has since said those results are outdated. Tests are presented as a way to define expected behavior, give agents a clear stopping point, and lock in verified behavior.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Nine fields to fill in an SEO content brief before writing a single word

A common scenario is described where a website owner gives a writer only a keyword and a word count, and the resulting article ranks nowhere three months after publishing despite being well written. The writer did nothing wrong; the brief simply left them to guess at intent, audience, and structure. The author rebuilds the brief as a checklist of nine fields, covering the primary and secondary keywords, search intent, the reader's goal, the page's role such as pillar or guide, required questions, supporting evidence, internal links, and post-publication KPIs. The fifth field, the URL's role, is the one most often overlooked, since two pages competing for the same intent end up cannibalizing each other. The article also lists technical checks such as a 200 status, self-referencing canonicals, indexable content in the initial HTML, and unique titles and meta descriptions. The author stresses that completing the form does not guarantee top rankings, but it removes guesswork and makes it possible to trace where a page went wrong.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Block Editor Shifts WordPress Plugin Development, But Classic Methods Remain

The WordPress Block Editor has changed how developers build features, letting users configure content elements visually instead of relying on shortcodes. Blocks are especially useful for page layouts, testimonials, and other reusable components that appear within content. However, the article argues that classic plugin development is not obsolete, since core functions like permissions, payments, database operations, and API integrations still run on server-side PHP. Many plugins therefore combine block interfaces with traditional backend logic. Overall, the Block Editor complements rather than replaces established WordPress development techniques.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Veteran developer says loyalty and quality lost out to a "good enough" market

A software developer with over 45 years of experience recounts starting on a Commodore 64 and progressing through assembly, the Amiga and C. For nearly 12 years he worked for a movie-industry client, often underbidding and working year-round without raising his fees. He says the client abruptly ended the relationship after deciding that "good enough" mattered more than being better. He reflects that underpricing and loyalty were mistakes in a market that rewards job-hopping, though these lessons conflict with his ethics. He concludes that his career may need to change if continued learning requires it.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Why Naive String Slicing Breaks Link Previews With Emoji and Urdu Text

Link preview titles and descriptions are often generated automatically from CMS fields, product feeds, or Markdown bodies rather than written by hand. A common one-line approach that slices strings to a fixed length introduces several bugs. JavaScript's slice counts UTF-16 code units, so it can split emoji such as flags or joined sequences, producing broken characters. The same issue affects Urdu, Arabic, and Hindi text, where combining marks can be detached from their base letters. The article proposes counting grapheme clusters with the built-in Intl.Segmenter, which reflects what readers perceive as single characters. It also offers a tested helper that cuts at word boundaries, falls back to grapheme cuts for overly long tokens, and includes the ellipsis within the length budget.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Lessons from running a 200-game browser arcade on instant-play design

A contributor who helps run PlayPokiGame, a free site of casual HTML5 games playable in-browser without downloads or signups, shares lessons from observing real player behavior. Time-to-interactive proved the strongest predictor of play sessions, with games showing a playable frame within about three seconds drawing far more engagement. The author also found that browser players skip tutorials, so controls should be shown in-game, and that short 10–20 minute play bursts favor games with natural 3–5 minute loops. Instant respawns, auto-pausing on tab blur, and periodic localStorage saves are highlighted as key to keeping players from abandoning games. The author argues these principles apply broadly to web products, given the browser runs on everything from cheap Chromebooks to high-end workstations.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Browser-based HEIC to JPG conversion for iPhone photos using JavaScript

iPhones save photos in HEIC format by default, which is smaller than JPG but often rejected by websites, forms, and Windows apps. Because only Safari can natively decode HEIC, the author recommends the heic2any library, which uses WebAssembly to convert files directly in the browser. The library is about 1.3 MB, so it is loaded only after a user selects a HEIC file to keep page loads fast. Developers are advised to handle multi-image HEIC outputs as arrays, check file extensions since MIME types can be empty, convert files sequentially to limit memory use, and use a quality setting of 0.9. Since the file never leaves the user's device, the approach also suits private photos.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Five On-Premises Project Management Tools Compared on Data Control and Workflow Fit

Organizations that must keep project records on infrastructure they manage need to weigh deployment options alongside approval steps, status reporting, and daily usability. A new practitioner's guide compares five tools on deployment and data control, workflow coverage, and operational fit. It recommends first confirming hosting requirements, then testing each candidate against a representative project from planning through reporting. The guide profiles ONES.com, Jira Data Center, Microsoft Project Server, Wrike, and OpenProject Enterprise, noting each one's strengths and caveats such as Jira Data Center's announced end-of-life timeline. Its suggested evaluation method is to trace a requirement through planned work and execution updates to see whether it surfaces in project reporting.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Treat LLM Agents Like New Bank Tellers: Five Controls That Transfer

A developer with a background in financial systems argues that LLM agents with tool access resemble new bank employees who are fast but cannot reliably distinguish genuine customers from fraudsters. The author suggests that long-established banking controls can be applied to these agents. Key recommendations include enforcing policy in code rather than in prompts, using a gate that checks each proposed tool call before execution, and treating retrieved document content as untrusted data to limit prompt injection. The piece also recommends detailed logging of every proposed tool call, including rejections, and using velocity checks to detect fraud patterns.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

New CLI Maps Broken Script Calls That Linters Miss

ShellCheck and Pylint check individual files but cannot see how scripts call one another, so a pipeline can pass linting and still fail. The author gives an example where a deploy script calls a file that was renamed earlier, leaving a stale reference nobody updated. The same gap affects AI-generated scripts that chain together through subprocess calls, sometimes forming loops or dead references. The author released script-chain-walk, a zero-dependency Python tool on PyPI that maps call relationships across shell and Python files, flagging missing targets and call cycles. Its --check flag exits with an error code when problems are found, making it usable in audits and continuous integration.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Oxlint and Oxfmt Cut a React Project's Lint Check From One Minute to Two Seconds

The Oxc project's linter Oxlint and formatter Oxfmt claim to run dramatically faster than ESLint and Prettier. To test this, the author timed both setups on Excalidraw, a 710-file open-source React and TypeScript codebase, on an ordinary laptop. The full check dropped from about 60 seconds to roughly 2 seconds, a real but smaller speedup of around 30x rather than the advertised 50 to 100x. Oxlint also reported 435 warnings where ESLint reported none, but most were stricter default rules rather than genuine bugs, though it did surface some useful React issues. The author notes that three custom Excalidraw rules have no direct Oxlint equivalent, and suggests the real work is reviewing which rules to keep.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Stakeholder Analysis for Software Projects: A Practical Template

A stakeholder analysis for a software project is a short table listing the people who can influence, block, use, or fund the product, along with their needs and level of influence. It is built early in discovery to plan interviews and sign-offs, and is updated until handover. The article explains that it serves three purposes: guiding interviews, identifying approval owners, and surfacing conflicts between stakeholders early. It groups relevant stakeholders into sponsors, decision makers, end users, support staff, integrated system owners, gatekeepers such as security and legal, delivery teams, and outside parties. It also recommends a concise set of columns, including power, interest, needs, and stance, plus a power-interest grid to prioritize engagement.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer releases open-source bash-write-targets tool to statically detect bash command file writes

A developer built a standalone, stdlib-only Python tool called bash-write-targets after their agent-guard project's README said such a tool did not yet exist. The tool statically parses a bash command string to list the files it would write, such as redirection targets and sed -i or mv destinations, and can output the results as JSON. It handles cases like heredocs and unresolvable variables, and it does not execute commands or use eval. The developer notes its limits: it cannot detect writes made inside interpreter code like python3 -c, nor network actions such as POSTing data. The project is MIT-licensed and available on GitHub and PyPI, with 19 smoke tests reported as passing.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Old Camera Phones Get No Annual Change Notice, So Check Five Settings Yearly

A phone kept running as a dedicated camera continues to receive platform changes with no notice listing what has altered, unlike Medicare plans, which send annual letters. The author recommends an annual review of five items that each take under a minute to check. One key risk is Android's app hibernation rule, which can reset camera permissions and clear cache for an app left unused, a common situation for a camera phone. Owners should also confirm that the two separate security-update dates still advance, and check whether automatic app updates are interrupting the camera app. Doing nothing, the piece argues, is not the same as standing still, since it means accepting whatever has changed.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Atlassian's AI agent to charge $1 per resolution in Customer Service Management from December 2026

Atlassian will begin billing $1.00 for each AI agent resolution in Customer Service Management from 3 December 2026, with no included allowance on any plan. The change prompted one team to delay its rollout, since it could not benchmark likely resolution volumes without testing in production. The article notes that the meter applies to Customer Service Management's AI agent, while Jira Service Management's self-service agent instead consumes Rovo credits, at 10 per reply in testing. Teams running Jira Service Management, particularly for internal employee requests, may therefore not be affected by the per-resolution charge. Whether the cost is worthwhile depends largely on how many resolved conversations would otherwise have reached a human agent.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

GoDaddy and Cloudflare Run About a Third of All .com DNS Names

An analysis of roughly 167 million .com names on 8 October 2026 found that GoDaddy hosts 21.66% of them and Cloudflare 12.94%, together accounting for about a third of the zone. The ranking groups each name under the company running its first nameserver, and the top 15 providers handle about 62% of names. Much of GoDaddy's lead comes from its size as a registrar, while many owners deliberately choose Cloudflare for speed and its free plan. Parking and marketplace services such as dns-parking.com and Afternic also appear high in the list, though their names often lack a working website. The authors note that concentration creates a single point of failure and recommend owners consider separating DNS from their registrar or host.

0 comments Read more at DEV Community
← NewerPage 195 of 2984Older →