GitHub Copilot Offers Built-In AI Agent Patterns Without Third-Party Frameworks

Developers can build effective AI agents using GitHub Copilot without relying on complex third-party frameworks, according to a guide published on DEV Community. Copilot natively supports key agentic design patterns including prompt chaining, routing, parallelization, and orchestrator-worker models directly within the IDE. Features such as Agent Mode, custom agents, prompt files, and Model Context Protocol (MCP) allow developers to integrate external systems and automate multi-step workflows. A central concept highlighted is avoiding 'context rot,' which occurs when a single prompt is overloaded with too many responsibilities like architecture, migration, and testing simultaneously. The guide recommends breaking tasks into modular, sequential steps with verification gates — such as running build or test commands — to improve accuracy and reliability.

Robotaxi Expansion Stalls Across US as Labor Groups and Lawmakers Push Back

Autonomous taxi services are growing in the United States, but regulatory and political resistance is mounting in several states. In New York, Governor Kathy Hochul withdrew a proposal to allow driverless robotaxis outside New York City following opposition from taxi drivers, unions, and lawmakers, leaving commercial driverless service still illegal statewide. In Washington DC, labor unions are actively opposing legislation to legalize commercial robotaxis, while lawmakers are considering a 200-vehicle cap and a 15-cent-per-mile fee to fund public transit upgrades and support displaced workers. The debates reflect a broader national tension between the rapid technological advancement of autonomous vehicles and the concerns of workers and regulators over safety, jobs, and oversight.

AI UI Testing Tools Score High in Demos but Struggle in Production, Builders Find

The team behind AI UI testing platform TestStar ran a single 19-step real-world test case eight times and recorded wildly inconsistent results, with passes and failures stemming from entirely different causes each run. A deeper investigation revealed that while AI execution engines performed their tasks correctly, an underlying infrastructure issue — sync memory writes backing up a worker thread by step 19 — caused the platform itself to collapse. The builders argue that popular AI browser-testing tools like Browser-Use, Skyvern, and Midscene function as execution engines only, lacking what they call a 'Harness layer' covering failure handling, self-healing, observability, idempotency, memory, scheduling, and page knowledge. Real-world debugging also exposed that self-healing logic must verify intended user outcomes rather than trusting system-reported success, after a CSS bug caused the AI to click an 'Archive' button it believed was 'Delete'. The team warns that teams evaluating AI testing tools should ask about the Harness layer upfront, noting one organisation spent nine months picking a tool and then building missing infrastructure themselves.

AI Tops YouTube as Developers' Most Popular Learning Tool, MLH Survey Finds

A new Season Census by Major League Hacking found that 75% of developers globally now use AI in some form to learn technical skills, surpassing YouTube at 71% and online courses at 59%. Rather than replacing other learning methods, AI learners actually engage more with additional channels — attending more events, watching more videos, and taking more courses than non-AI learners. The survey also found that developers who learn via AI are roughly a third more likely to adopt tools like Firebase and Supabase, suggesting that LLMs may be steering learners toward certain technology stacks. Adding to the shift, documentation platform Mintlify reports that two-thirds of its hosted doc visits now come from AI agents rather than humans, up from just 15% earlier this year. These trends signal that developer tool companies must now optimize their content and messaging for both human developers and the AI models those developers learn from.

Nvidia Senior Manager Indicted Over Alleged AI Server Smuggling to China

A senior Nvidia manager has been indicted in connection with a scheme to illegally smuggle AI servers to China. The case is linked to server manufacturer Supermicro, which had previously drawn criticism from Nvidia CEO Jensen Huang over the smuggling activity. The indictment suggests the illegal export operation involved coordination between individuals at multiple companies. The case highlights growing concerns around the circumvention of US export controls on advanced AI hardware destined for China.

Google Cloud Course Breaks Down How AI Agents Work and Why They Matter

A developer enrolled in Google Cloud's Agentic Summer course, powered by the Gemini Enterprise Agent Ready (GEAR) program, is sharing structured notes on how AI agents function. An AI agent is defined as a software system that uses a large language model (LLM) to accomplish goals on a user's behalf, built around three core components: an LLM, tools, and contextual memory. Tools are often accessed via the Model Context Protocol (MCP), which goes beyond standard APIs by providing agents with guidance on when and how to use specific integrations. Memory management is highlighted as critical, involving strategies like context compression, summarization, and selective retention to keep agents focused and effective. The course also introduces agent design patterns, ranging from simple single-agent setups to more complex multi-agent architectures suited for advanced tasks.

No alliance with SAD? BL Santosh says BJP to go solo in Punjab polls

Why Your App Fails SSL Checks That Browsers Pass — and How to Fix It

When a browser shows a green padlock but server-side tools like curl, Node.js, Python, or Go throw an SSL certificate error, the root cause is almost always a misconfigured server, not a strict client. TLS servers are required to send the full certificate chain — the leaf certificate plus all intermediate certificates — but many are deployed with only the leaf, leaving API clients unable to verify the path to a trusted root. Browsers mask this flaw by caching intermediates from previous visits and using Authority Information Access (AIA) fetching to download missing ones, which most server-side stacks deliberately avoid. The most common trigger is using Let's Encrypt's cert.pem instead of fullchain.pem, which omits the necessary intermediate certificate. Disabling SSL verification with flags like curl -k or rejectUnauthorized: false is not a fix — it silences a warning that correctly identifies a real server misconfiguration.

Solo dev discovers all 155 tests passed while every real AI request silently failed

A solo developer building GigLegal, a contract tool for freelancers on Next.js 15, found that all 155 Playwright end-to-end tests were passing while every real AI request was failing in production. The root cause was a provider resolver that automatically switched to a mock AI service during tests, meaning the test suite never interacted with the actual API. Behind this masking issue lay three stacked bugs: an invalid structured output configuration, a schema with 66 optional fields that exceeded the vendor's undocumented 24-field limit, and a timeout problem on a 20-field policy step. Splitting the large single AI call into smaller, focused requests ultimately resolved the failures, with the developer noting that mocks reflect assumptions rather than real API behavior.

Iran's IRGC-Linked Media Posts Rs 95 Crore Bounty on Trump's Son Barron

An IRGC-affiliated Iranian media outlet has released a video targeting Barron Trump, the 20-year-old son of US President Donald Trump. The video, titled 'Where to Kill Barron Trump,' was broadcast on Channel 3 of Iranian state television. It reportedly tracks Barron's movements, raising serious security concerns. A bounty of approximately Rs 95 crore has been announced in connection with the threat. The development marks a significant escalation in tensions between Iran and the Trump administration.

Developer builds 'ndo' CLI tool to fix frustrations with the 'just' command runner

A developer frustrated with recurring limitations in the popular 'just' command runner has released an alternative CLI tool called 'ndo', currently at version 1.1.0. The tool allows users to add command shortcuts directly from the terminal without editing a configuration file, addressing a key friction point with 'just'. Unlike 'just', ndo supports both a local project-level recipe file and a central file shared across all projects on a machine, with local recipes taking silent precedence when conflicts arise. It also simplifies argument passing at the command line and includes a variable lookup table for aliasing frequently used folder paths. The single static binary supports Linux, macOS, and Windows on amd64 and arm64 architectures, uses TOML for configuration, and is available via Homebrew, Scoop, or a curl install script.

Window of Patna-Gomti Nagar Vande Bharat damaged in stone-pelting incident in Bihar

A window of the Patna-Gomti Nagar Vande Bharat Express was damaged after miscreants pelted stones at the train in Bihar. The incident prompted railway authorities to file a formal case in connection with the attack. Stone-pelting on trains has been a recurring concern for Indian Railways, particularly affecting the Vande Bharat fleet. Authorities are investigating the matter to identify those responsible for the damage.

Proposal to hold joint polls flawed, undermines basic structure of Constitution, says Chidambaram

Trump tightens Iran squeeze, hurting China and India in the process

Developer Releases Modern GUI Library for Ada with CSS and XML Support

A developer has open-sourced a new graphical user interface library called Adi2, designed for the Ada programming language. The project, shared on GitHub, brings modern UI capabilities to Ada by supporting CSS-based styling and XML-defined user interfaces. It also integrates with SDL3, a widely used multimedia and graphics framework. The library was posted to Hacker News under the 'Show HN' category, attracting 26 upvotes and 14 comments from the community. The release aims to modernize Ada's GUI development tooling, which has historically lagged behind more mainstream languages.

Public services are increasingly strained by LLM-written appeals for benefits

Article URL: https://arxiv.org/abs/2608.16603 Comments URL: https://news.ycombinator.com/item?id=49422227 Points: 51 # Comments: 60

Ox-Alpha Is GLM?

Article URL: https://dejan.ai/blog/ox-alpha/ Comments URL: https://news.ycombinator.com/item?id=49422226 Points: 30 # Comments: 8

How WWE Star Cody Rhodes Built a Luxury Lifestyle Beyond the Ring

WWE champion Cody Rhodes has translated his professional wrestling success into a high-end personal lifestyle. Known by his ring name 'The American Nightmare,' Rhodes has developed a distinct public image encompassing upscale homes, a collection of cars, custom suits, and luxury watches. His financial achievements in wrestling appear to underpin both his material acquisitions and his disciplined daily routine. Rhodes has also been expanding his presence beyond sports entertainment through a growing business profile.

A Blackstone real estate company exposed SSN digits, DOBs, addresses and more

Article URL: https://alexschapiro.com/security/vulnerability/2026/07/16/beam-living-graphql-data-exposure Comments URL: https://news.ycombinator.com/item?id=49422204 Points: 76 # Comments: 22

England drop Brydon Carse for Pakistan 2nd Test after handcuff video surfaces

England have dropped fast bowler Brydon Carse from their squad for the second Test against Pakistan following the emergence of footage showing him in handcuffs outside a Derby nightclub. The Cricket Regulator has launched an investigation into the incident. Carse was reportedly released without charge, with the episode said to have occurred during celebrations of Durham's County Championship victory. The incident adds to a series of recent disciplinary concerns surrounding England players linked to nightclub-related behaviour.