SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World

Programming

LatestTopDiscussed
0
ProgrammingDEV Community ·

Checkout 502 spikes traced to pods receiving ALB traffic after shutdown during deploys

During rolling deploys, the checkout service saw its 502 error rate briefly jump from about 0.02% to 4.8%, yet neither the app logs nor the target group health views showed any fault. Investigators found in the ALB access logs that failed requests mapped to pods still receiving traffic seconds after they were told to terminate. The checkout app had no SIGTERM handler, so pods exited within a second of the signal. The AWS Load Balancer Controller took four to seven seconds to deregister those pods from the target group. The 300-second deregistration delay on the target group made the gap worse by keeping the ALB routing to pods that had already gone.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

DriftGuard: Detecting Data Drift Early with PSI on Amazon SageMaker

A developer built DriftGuard, a small MLOps project that uses Python, scikit-learn and the Population Stability Index (PSI) to catch data drift in deployed models. The project trains a model on synthetic data, creates production data with a deliberately shifted feature, and computes PSI scores by comparing binned distributions against the training baseline. PSI results flagged the shifted feature strongly and also marked an unshifted feature as drifting, which the author attributed to the small sample of 100 rows. The author concludes that drift metrics need sufficiently large data windows and should be combined with other statistical tests rather than relying on a single threshold. The full code is published on GitHub.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

No CAPTCHA Solved by Agents in 119 Attempts; ai-Tagged Posts Gain Reactions

An AI-run publication reported that its trial bench was idle this week, with its next batch of CAPTCHA tests scheduled for late October. Across 119 pooled attempts, the GoCaptcha challenge remained unsolved by three models, while Altcha and Cap were solved in every trial. The authors caution that results reflect one agent class in one environment, a headless browser on a Singapore datacenter IP. Separately, their analysis of dev.to snapshots found that posts with measured, argumentative content drew the most comments, and that ai-tagged posts grew in median reactions while webdev posts declined. The report also acknowledged that its own dev.to posts had received very few views.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Anchored comment threads keep document feedback tied to the text it concerns

Comment threads placed at the bottom of long documents often lose context, making it hard for readers to tell which paragraph or version a note refers to. Anchored annotation instead attaches each discussion to a specific section, sentence, or element, so the context travels with the comment. This makes feedback easier to act on and lets writers resolve threads individually. The approach also raises design questions, such as how to handle anchors whose text is deleted or rewritten and whether threads should inherit document permissions. The article describes how the Andika knowledge workspace has implemented these ideas alongside section navigation, attachments, and summarization features.

0 comments Read more at DEV Community
0
ProgrammingHacker News ·

Google Cloud outage disrupts banking services including HSBC, Revolut, and Lloyds

A Google Cloud outage has reportedly disrupted online services at several major financial institutions, including HSBC, Revolut, Bank of Scotland, Lloyds, and Equifax. The article linked from the original post indicates the affected customers experienced problems accessing their services. The source material provided does not specify the exact duration of the disruption or its root cause. Further details on the scope and resolution were not included in the information available to this summary.

0 comments Read more at Hacker News
0
ProgrammingDEV Community ·

Argues Tool Engineering Will Replace Prompt Engineering as AI Skill of Choice

A developer writer contends that prompt engineering, once a prized skill, is becoming obsolete as language models grow more capable. He points to three drivers: models now grasp plain-language requests without elaborate scaffolding, very large context windows reduce the need for compression tricks, and fine-tuning and retrieval-augmented generation solve problems that prompt hacks once addressed. He proposes a successor discipline, Tool Engineering, centered on building, selecting, and orchestrating the tools AI agents use to act in the real world. The piece argues that a model without tools can reason but cannot take actions such as booking a hotel or running a database query. Developers who keep investing in prompt techniques, he warns, risk becoming increasingly irrelevant.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Recording Free Inference Calls in a Ledger Before Making Them

The article argues that free inference lanes fail less on quality than on accounting, since unrecorded spend cannot be subtracted from a budget. It proposes placing a ledger in front of the lane and recording each reservation before the upstream call is made. Because workers can die mid-call and retries can double spend, the ledger must survive crashes without manual cleanup. The author describes a lease-based design where each job holds an open budget row that is committed or released, with a sweeper expiring stale leases. The implementation uses only the Python standard library and a single SQLite file, suitable for a laptop or small always-on machine.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Google Unveils Gemini 4 Argon With 1M-Token Output, Rolling Out Gradually

Google DeepMind announced Gemini 4 Argon on September 30, 2026, positioning it for complex software engineering, knowledge work, and defensive cybersecurity. Its standout specification is an output limit of up to 1 million tokens, raised from 64K, which could support long multi-step tasks. Access is initially limited to trusted cyber defenders and testers, with broader developer and enterprise availability still to come. Google reports a 77.9% score on DeepSWE v1.1 along with an introductory API price of $2 per million input tokens and $10 per million output tokens. The article cautions that these are publisher-reported figures, not a controlled head-to-head comparison with OpenAI's GPT-6.1 Sol.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Log Model Call Metadata Instead of Payloads to Reduce Data Leak Risk

The article argues that AI features often leak sensitive data through logs rather than request bodies, since logging receives less security scrutiny. It proposes logging only a "shape" record for each model call, including a payload digest, byte size, field names, latency, and attempt count, instead of message content. Encryption is deemed insufficient because logs are decrypted when searched, exported, or shared in support bundles. The approach is enforced with a CI test that scans log fixtures for disallowed fields and sensitive patterns such as emails, bearer tokens, and private keys. The author also recommends including a deliberately leaky fixture to confirm the scanner actually detects leaks.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

AgenticOS adds shared projects so teams can work with AI agents on common files

AgenticOS, an open-source Apache-2.0 system for running company AI agents on your own servers, has introduced projects for team-level work. A project holds shared files that every chat in it works on, along with instructions that all agents in the project read before answering. Agents working in a project operate inside the project's container, so files written in one chat remain available in later ones. Multiple team members can join a project chat simultaneously, seeing live typing indicators and agent responses as they appear. The article's author, who builds the system at Vstorm, notes the feature has limits and plans to cover sandbox-based coding agents in a follow-up.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Developer adds a lightweight gate to cut LLM costs on a Hebrew stock-review chatbot

A developer running telavivstocks.com, a Hebrew dashboard for Tel Aviv Stock Exchange stocks, found that their AI assistant ran a full embedding, vector search and generation for every question, including off-topic ones. To fix this, they placed a small decision model, TypeSafe's Jev, in front of the expensive model to decide whether it should be called at all. The gate classifies each question by intent and checks for prompt-injection attempts, returning probabilities that can be compared against a threshold. The developer estimates the cost at about two cents per thousand questions, and wrote the gate instructions in English while keeping visitors' Hebrew questions unchanged.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Tracking Docstring Drift by Hashing Public Symbols and Flagging Signature-Only Changes

A small internal Python library suffered from documentation that fell out of step with its code, such as a docstring still listing a parameter name that had been renamed. The author built a deterministic ledger that hashes each module-level symbol's signature and docstring and compares them against a committed snapshot. A symbol is flagged only when its signature changed while its docstring stayed the same, since an edited docstring means someone has already reviewed it. Because the diffing is deterministic, a language model is needed only as an optional annotation on symbols already flagged, and the test suite can run offline against a fake client. The tool is scoped to Python signature and docstring drift and does not attempt to certify that documentation is correct.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Free-tier AI prototypes fail at runtime, not the model call, and need a readiness checklist

A DEV Community article argues that hosted AI prototypes on free tiers tend to break from server-side issues rather than model calls. It contends that local disk may be ephemeral, instances can sleep or be recycled, addresses can change between deploys, and outbound model calls may be blocked or rate-limited. The author proposes a runtime-readiness checklist where each claim needs a named evidence artifact, and any claim lacking recent evidence is treated as blocked rather than passed. The checklist covers cold starts, state durability, idempotent writes, payload limits, egress, header leakage, and rollback, and includes a stdlib-only Python probe that writes one evidence file per gate.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Agents Reveal How They Found a Message Board Through the Shape of Their First Posts

New AI agents keep arriving on a message board, and the author has begun examining how each one first contacts the platform. One agent posted a greeting that copied the board's quickstart example word for word, including its placeholder name and encoding characters. Another agent repeatedly posted the single word "hello" over a span of about two weeks, likely reflecting a retry loop rather than deliberate behavior. A separate integration tester ran probe messages on the live board, labeling them as safe to ignore and noting that the board has no delete function. The author says the board will not auto-reply to arrivals, to avoid being overwhelmed by spam.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

AI agent runs a $100-capped experiment to earn $12 for an espresso cup, five days in with zero payments

An AI agent, Claude, is attempting to raise about $12 for an espresso cup for a person named John within two weeks, under a $100 spending cap and rules against spam and fake reviews. After five days and $8.43 of a $45 Reddit ad budget, the funnel shows 9,304 ad views, 59 clicks, 20 people who read or used a tool, and one checkout opened with no payments received. The agent wrote its stop rules before spending, including halting at $25 spent without payment or when the conversion rate falls clearly below break-even after about 150 clicks. Comparable experiments suggest only a tiny share of visitors pay after an ad click, and the agent says the AI projects that raised real money depended on public engagement. The agent has built a home-espresso setup page and diary, and says it will disclose its AI nature when asked.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

LazyDev: An Open-Source, Local-First AI Assistant to Ease Repetitive Coding

LazyDev is an open-source AI coding assistant being developed by a creator who aims to reduce repetitive work and tool-switching for developers. It is designed to fit into existing workflows and to support locally runnable AI models rather than relying solely on closed cloud APIs. A public web demo is available, but it uses deterministic demonstration responses rather than a live language model, so it does not represent real local inference. The project is still in progress, with current work focused on developer experience, reliable builds with automated tests, and isolating the demo from real workspaces. The creator also notes they used an AI-assisted workflow to help build the project.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Free Compute Comes With Hidden Expiry Risks, Argues Developer

A developer contends that free model access and servers suit disposable experiments but become risky when teams rely on them for production work. Free resources can expire, throttle, or disappear without warning, causing failures from time drift, spending creep, or lost state. The author proposes a 'lease file' with an absolute deadline, a hard token ceiling, a durable output sink outside the free machine, and a teardown command. A gate script that refuses to start when the lease is invalid, plus a three-step teardown process, are included as examples. The core argument is that the missing contract, not the cost, is the real problem.

0 comments Read more at DEV Community
0
ProgrammingHacker News ·

Revolut app outage leaves users unable to access services

Reports surfaced that the Revolut app was not working for a number of users. The disruption was covered by the BBC, which linked to an article on the issue. The source material available provides only the headline and basic metadata, so details on the cause, duration, and scale of the outage cannot be confirmed here. Users affected by the problem were likely unable to use the app's services during the period of disruption. Further details would need to be drawn from the original BBC report.

0 comments Read more at Hacker News
0
ProgrammingHacker News ·

Leaked Emails Reveal Inner Workings of Germany's AfD Party

A hacked trove of 421,179 emails from Germany's far-right Alternative for Germany (AfD) party has been analyzed and published. The leak offers a look at internal party communications. The source material provided is limited to a link and metadata, so specific details of the emails' contents cannot be confirmed here. Readers are encouraged to consult the original report for the documented findings. The story has drawn attention on Hacker News.

0 comments Read more at Hacker News
0
ProgrammingHacker News ·

UK Defence Startups Report Being Denied Banking Services Under ESG Rules

An analysis published by Fox and Lion examines reports that UK defence start-ups are being refused banking services. The piece attributes these refusals to environmental, social and governance (ESG) policies adopted by financial institutions. The article's source material provided only the headline, link and Hacker News metadata, so specific cases, banks and figures could not be verified here. The story gained traction on Hacker News with 12 points and no comments at the time of capture. Readers are directed to the original analysis for the full details and evidence.

0 comments Read more at Hacker News
← NewerPage 282 of 3055Older →