Alvaro Lopez Ortega / 2026-08-03 Briefing

Created Mon, 03 Aug 2026 19:15:19 +0000 Modified Tue, 04 Aug 2026 19:15:32 +0000
8711 Words

Hughesnet filed for Chapter 11 after losing half its subscribers to Starlink, refocusing on government clients. Microsoft plans to bring Xbox 360 games to PC and next-gen consoles by 2027. The White House quietly met its AI evaluation deadline; OpenAI, Google, and Anthropic staffers were invited to review. Apple is legally challenging the UK’s encrypted-data backdoor demand. Visa bought AI fraud-detection firm BioCatch for $2.4B.

🤖 AI & Automation

Alibaba prices Qwen3.8-Max API at $2/$6 per 1M tokens, undercutting Kimi K3’s $3/$15.

Apple sued former employees now at OpenAI, accusing them of stealing trade secrets by improperly accessing confidential Apple documents after leaving. Before the lawsuit, Apple lawyers sent letters to over 400 former Apple employees at OpenAI, stating they possessed confidential material. The article also examines how an Apple iCloud policy contributed to the leaks ahead of the suit.

Indeed: UK employers create senior software/IT jobs as AI enhances experience value, cut elsewhere

According to Indeed data, UK job postings have fallen about 10% since January 2025, but this masks a growing two-track labor market. Employers are creating jobs for experienced staff with AI skills while cutting vacancies elsewhere, as firms adopt AI to cut costs and boost productivity.

What’s the largest software project AI can complete on its own?

MirrorCode, a benchmark co-developed with METR, tests AI models on long-horizon coding tasks by requiring them to reimplement entire programs end-to-end without seeing the original source code, matching exact outputs on held-out tests. It includes 25 target programs across computing areas and is designed to be scale-aware and cheat-resistant, though data contamination from pretraining remains a caveat. AI models like Claude Opus 4.7 have already solved some tasks, with open-source code and a leaderboard tracking progress.

MiniMax H3 Day-0 Support in ComfyUI: Open Weights, Native Audio, and 2K Video

MiniMax H3, an open-weights third-generation video model, launched with native day-zero support in ComfyUI. It generates video clips up to 2K resolution and 15 seconds long from text, images, video, or audio, producing native stereo audio in the same pass. The model supports multimodal context understanding, reference-to-video, motion transfer, and first-and-last-frame control for editing tasks.

AirLLM 70B inference with single 4GB GPU

AirLLM cuts inference memory usage so dramatically that 70B models can run on a single 4GB GPU without quantization, distillation, or pruning. By streaming sparse MoE models one expert at a time, it can run 405B Llama 3.1 on 8GB, DeepSeek-V3 (671B) on ~12GB, and the 2.8T-parameter Kimi K3 on under 4GB VRAM.

1 in 4 people in Japan believes AI could replace friends and family: poll

A Jiji survey found that 24.9% of Japanese people believe advanced AI will eventually replace their friends and family, while 65.1% disagree. Younger respondents were more likely to hold this belief. Separately, 30.8% of respondents use generative AI, primarily for work.

The Shape of Things to Come

Reusable coding agent harnesses are failing and becoming obsolete, as they must be bespoke and integrated into the application itself; the author abandoned his own reusable harness, Gas Town, due to model quirks. He now uses the Claude Fable model to make significant progress on his long-neglected MMO Wyvern, clearing over half of a 100-year backlog, though content creation and taste remain major bottlenecks. He is building a new bespoke harness, Wheelhouse, which he believes reflects the future of agent orchestration.

The AI Productivity Gap

AI’s main benefit is accelerating code writing, which is only a small portion of developers’ overall work, yielding modest total productivity gains of about 15% for senior engineers and 25% for junior engineers. Junior developers see larger gains because they spend more of their time coding, so the idea that AI makes them obsolete is misguided. Most engineering tasks—such as design, debugging, reviews, and meetings—remain largely unaffected by AI.

AI migrated legacy COBOL programs to Java, bugs included

The paper introduces “Locksmith Loop,” an agentic test-synthesis method that validates COBOL-to-Java migrations by iteratively searching input mocks to penetrate branches and applying parity-preserving mutations. Tested on three case studies, it improved branch coverage past input-search plateaus, achieving near-complete coverage on open-source programs and 91.90% on an internal production-like COBOL program, with generated Java matching COBOL under deterministic parity checks.

Qwen 3.8 Max Live Now

Qwen3.8-Max, a 2.4-trillion-parameter MoE model, is now live, delivering major advances in coding and professional work with autonomous, end-to-end project completion across legal, financial, and design domains. It also supports native visual understanding for deep analysis of long documents and videos, and uses closed feedback loops to plan and iterate on long-horizon tasks.

Categorization with NLP

The author describes a manually crafted grocery categorization algorithm that lexes and stems product names into sorted word lists, then matches unigrams in a priority-ordered CSV to classify derived products like “apple juice” as a drink rather than produce. This rule-based approach avoids machine learning due to insufficient data and relies on the significance of word order in the database.

Don’t be a meat proxy

The article criticizes people who relay AI-generated responses verbatim, calling them “meat proxies” that add no value since recipients can query the AI directly. It advises users to read, understand, and rewrite AI outputs in their own words, especially in code reviews, where blindly passing AI responses shifts the real work onto human reviewers.

Google Earth’s AI makeover survives one trip around the Sun

A year after Google Earth introduced an AI-powered feature that let users rewrite the planet with text prompts, the feature has survived its first year, but the experiment exposed notable drawbacks. The tool’s open-ended nature led to unintended consequences, highlighting the challenges of allowing anyone to alter the map with a prompt.

Claude Code is revolutionizing digital archaeology. Enterprise better dig it

Claude Code, an AI coding tool, is proving highly effective in digital archaeology by helping reverse-engineer vintage systems, surprising even skeptics. Notable projects include designing an accelerator for Acorn Archimedes to revive Inmos Transputers and fully disassembling Acorn’s 6502-based Network Filing System. These efforts demonstrate LLMs’ unexpected prowess in reconstructing and documenting obscure computing history.

🔐 Security & Privacy

Apple has filed a new legal challenge against the UK government’s demand for a backdoor to access encrypted customer data, according to a UK court filing from July. The iPhone maker is contesting the latest government order that would require allowing access to encrypted user data.

Analysis: 50+ US officers accused of misusing license-plate readers, including stalking

According to a Washington Post analysis of police and court records, at least 50 law enforcement officers have been charged or accused of misusing license-plate reader systems such as Flock’s, including for stalking. The nationwide camera network, built to fight crime, was reportedly exploited by rogue officers to track individuals for personal reasons.

Leaked Flock Guide Shows How It Teaches Cops to Promote Its Tech

A leaked Flock guide obtained by 404 Media coaches police on how to pitch its license plate reader technology to city councils, advising them to “own the narrative” early, avoid debating mass surveillance claims, and use a scripted presentation. The guide comes amid controversies over Flock’s data sharing with ICE, its use in abortion-related searches, and surveillance of protesters, as cities reconsider contracts.

Flock’s CEO Faced Me After Its Cameras Led to My Wrongful Stop

The article promotes a Drivecast episode in which the host interviews Flock Safety CEO Garrett Langley about a wrongful police stop involving Flock’s license plate cameras. They discuss public backlash over privacy and Fourth Amendment concerns, calls for regulation, and Langley’s past remark equating vandalism of Flock equipment to domestic terrorism.

Why does Mail app contact iCloud when sending a non-iCloud email?

Despite having iCloud and Contacts disabled in Mail settings, sending a non-iCloud email triggers iCloud connections via cloudd and nsurlsessiond to gateway.icloud.com. Logs indicate the app attempts to sync the “Previous Recipients” list (com.apple.mail.recents) to CloudKit, even though Contacts sync is turned off. The author plans to block these connections with Little Snitch rules.

SPF Record Syntax: Mechanisms, Qualifiers, Modifiers, and Macros

An SPF record is a single TXT record with strict syntax—a version tag, mechanisms, and modifiers—and any syntax error invalidates the entire record, causing authentication to fail. Eight mechanisms test the client IP, including include, which recursively evaluates another domain’s SPF record and matches only on a Pass result. In practice, most records rely on include, ip4, ip6, and all, and mechanisms listed after “all” are ignored.

Critical CVE issued for hallucinated SQLite vulnerability

A new GitHub repository published dozens of SQLite CVE advisories that appear to be AI-generated, yet NVD and CISA initially flagged many as critical. JFrog’s audit found the cited functions and fixes often don’t exist, PoCs trigger no crashes, and the issues aren’t in SQLite’s official advisory list. One CVE was downgraded from 10.0 to 7.6 after review, underscoring the unreliability of these automated reports.

ICE Collected Nearly 1M People’s DNA Last Year–Including Young Children

ICE’s expanded DNA collection from immigration detainees has made the agency the largest single source of new genetic profiles in the FBI’s criminal database, potentially adding around 920,000 profiles in 2025—mostly from people without criminal convictions, including families and children. The campaign was highlighted by the case of Hugo Moreno-Mendez, who was arrested at a probation check-in, prosecuted for refusing to provide a DNA sample, and found guilty, underscoring the threat of prosecution used to compel compliance.

Nightcrawler – A local AI pentesting agent running on a smartphone

Nightcrawler is an autonomous penetration testing agent that runs entirely on a smartphone, using a local 1.2B-parameter AI model on the device’s GPU to guide actions without cloud connectivity. It discovers hosts, maps services, tests for vulnerabilities and default credentials, and generates a pentest report while operating stealthily by rotating across hosts with small actions over hours. Optionally, it can crack WPA2 WiFi networks when deployed without existing network access.

What DMARC Protects You From, and What It Does Not

DMARC does not stop phishing or spam outright; it verifies whether the domain in the visible From line authorized the message through SPF or DKIM alignment. It passes if either SPF or DKIM authenticates and aligns with the visible sender, and fails only when neither does. SPF checks the hidden envelope address, DKIM signs the message, and DMARC ties those results to the sender address users actually see.

PISIGuard: Protect your personal and sensitive info when you chat with AI

PISIGuard is a browser extension that automatically detects and masks personal or sensitive information—such as names, emails, phone numbers, and API keys—in AI chat messages, replacing them with placeholders and restoring the original values in replies. All processing happens locally in the browser, with no data sent to external services. It supports platforms like ChatGPT, Claude, and DeepSeek, and is available as a Firefox add-on and via GitHub releases.

How rogue officers turned a nationwide camera network into a tool for stalking

A nationwide license-plate camera network by Flock, designed for crime fighting, was misused by at least 50 law enforcement officers accused of stalking or harassing individuals. One victim, Marci Bakely, said her ex-boyfriend seemed to always know where she was, highlighting the abuse of the system.

Sony BMG copy protection rootkit scandal from 2005

In 2005, Sony BMG was found to have secretly installed copy-protection software on about 22 million CDs that acted as rootkits, modified users’ operating systems, and created security vulnerabilities exploited by malware. The company initially denied harm, but its official uninstaller introduced further flaws, sparking public outcry, lawsuits, and government investigations. Sony BMG eventually settled with consumers, recalled about 10% of the affected CDs, and suspended CD copy protection in early 2007.

Apple engineer says he was fired after refusing to send cust. device IDs to AT&T

Former Apple engineer Toby Boardman alleges he was fired for refusing to share iPhone IMEI numbers with AT&T without customer releases, which he claims violated Apple’s privacy policies. Apple denies the claim, citing performance reasons, and a jury trial is scheduled for November 2027.

RFC 9851: TLS 1.2 is in Feature Freeze

RFC 9851 places TLS 1.2 in a feature freeze, approving no new changes except urgent security fixes, new TLS Exporter Labels, and new ALPN Protocol IDs. The freeze applies only to TLS, not DTLS, and post-quantum cryptography will not be specified for TLS 1.2, encouraging migration to TLS 1.3.

SQLite Critical CVEs or LLM Slop?

A batch of SQLite CVEs published in a new GitHub repo was initially marked critical by NVD and CISA, but JFrog found the advisories were AI-generated “LLM slop” with nonexistent code, broken PoCs, and no official SQLite listing. The cited functions and bug mechanics didn’t match the real source code, and the CVSS scores were later downgraded.

Polish retail giant Żabka breached via external service provider: data up for sale

Polish retail giant Żabka confirmed a cyberattack via an external service provider, blocking unauthorized access and notifying Poland’s data regulator while stating transactions and business operations remain secure. A threat actor is selling allegedly stolen data for €5,000, claiming it includes 541,000 Jira issues, source code, employee records, and production credentials. Researchers warn the exposed details could enable further attacks.

Apple fights surge of questionable vulnerability reports

Apple has limited security researchers’ vulnerability submissions and introduced a 30-day waiting period after a surge of low-quality, AI-generated reports. An Italian firm found over 50 MacBook bugs using ChatGPT, including a critical exploit, but was initially blocked from reporting it. Apple insists humans review all reports, but researchers worry the limits could delay fixes for serious flaws.

Russian spies turn public Wi-Fi into malware delivery systems

Russian spies are exploiting public Wi-Fi networks to deliver malware, with a focus on captive portals. Their methods include keyloggers, audio-visual surveillance, and token theft. The hospitality sector has been put on alert.

The Police National Legal Database (PNLD) confirmed it suffered a data theft after a dark web leak. The threat actor ExfilSquad claimed to have stolen 135,000 contact records, just weeks after the group targeted the Department for Education.

Water system cyberattacks spread to Georgia, Michigan amid US-Iran conflict

Water system cyberattacks have now spread to Georgia and Michigan, occurring amid the ongoing US-Iran conflict. President Trump rejected the theory that Tehran was responsible, instead blaming Minnesota’s governor, whom he called “grossly incompetent.”

UK government investment arm cops to 40-hour leak of officials’ contact details

The UK government’s investment arm admitted to a 40-hour data leak that exposed officials’ contact details. The breach occurred because an employee failed to follow security policy, leaving an internal management file publicly accessible.

AI is ‘both the weapon and the target’ in latest wave of cyberattacks

CrowdStrike reports an 89% surge in machine-assisted cyberattack activity, with AI serving as both the primary weapon and target. Attackers are exploiting AI systems while also using AI to accelerate attacks, forcing organizations to shrink patch windows to just 48 hours.

IT boss left root session open for bring-your-kid-to-work day

An IT administrator left a root session open during bring-your-kid-to-work day, allowing a child to type a Unix command. The incident serves as a cautionary reminder about the risks of leaving privileged terminals accessible.

💰 Business & Funding

Hughesnet has filed for Chapter 11 bankruptcy after losing over half its subscribers to Starlink and running low on cash. The company will continue serving customers during restructuring while shifting focus to enterprise, government, and defense clients, and plans to lay off 400 employees.

Valar raises $1B Series B led by Sequoia at $6B valuation for nuclear-powered data centers

Sequoia Capital led a $1 billion Series B round for nuclear startup Valar Atomics, valuing the company at $6 billion post-money. The funding will help Valar move from demonstrating small modular reactors to volume production, and the company also secured a $200 million credit facility.

Mariana Minerals, which builds/operates mines with MarianaOS, raises $310M Series B (total $400M)

Mariana Minerals, a software-focused mining startup, raised a $310 million Series B led by Khosla Ventures, bringing its total funding to $400 million at a $1.5 billion valuation. The company aims to lower costs for critical metals like lithium, copper, and aluminum, which it says are essential for the modern economy amid China’s dominance in global mineral processing.

AI security platform Zenity raises $125M Series C led by Norwest; total ~$185M

Zenity, an Israeli cybersecurity startup specializing in securing AI agents, raised $125 million in a Series C round led by Norwest Venture Partners, bringing its total funding to about $185 million. The company, which generates tens of millions in annual revenue and has tripled it for two consecutive years, plans to accelerate growth amid rising enterprise demand for AI agent security.

Visa to acquire AI fraud detection firm BioCatch for $2.4B cash

Visa has agreed to acquire Israeli fraud-prevention company BioCatch for $2.4 billion in cash. BioCatch uses artificial intelligence to detect fraud and distinguish legitimate users from attackers in real time. The company is being bought from Permira and other shareholders.

Horizon3 raises $250M Series E at $2B valuation

Horizon3, a cybersecurity startup using its AI-powered NodeZero platform for autonomous penetration testing, raised $250 million in a Series E round at a $2 billion valuation, more than tripling its valuation in 14 months. The funding, backed by returning investors NightDragon and NEA, comes amid rising enterprise demand for continuous, large-scale security testing as AI-driven attacks accelerate. The company reported nearly $100 million in annual recurring revenue and 120% year-over-year growth.

Saylor’s Strategy sells 1,638 BTC for ~$104.7M; holdings at 842,138 (~4% supply), $10.9B paper loss

Michael Saylor’s Strategy sold 1,638 bitcoin for approximately $104.7 million, reducing its total holdings to 842,138 BTC, or about 4% of bitcoin’s total supply. The proceeds were used to fund preferred stock distributions and repurchases of STRC, though the holdings now carry roughly $10.9 billion in paper losses at current prices.

June emerges from stealth with $20M from Marc Benioff’s Time Ventures for enterprise AI deployment

June, founded by former Salesforce executives, emerged from stealth with $20M in pre-seed funding led by Marc Benioff’s Time Ventures. The startup’s platform scans a company’s existing systems to identify bottlenecks and generate step-by-step plans for deploying AI agents, aiming to reduce reliance on forward-deployed engineers. It seeks to solve the challenge of integrating AI with complex legacy data platforms.

S. Korean AI chip firm DeepX raises $29M at $2.2B, in talks for $209M at $2.4B

South Korean AI chip designer DeepX raised about $29 million in the first tranche of its Series D round, valuing the company at roughly $2.2 billion—about four times its previous valuation. The funding came from existing investors BNW Investment and DS Asset Management, and the company is also in talks to raise approximately $209 million at a $2.4 billion valuation by September.

London AI chip startup Olix raises $312M at $3.3B valuation, up from $1B+, led by Fundomo with Arm

London-based AI chip startup Olix raised $312 million in funding led by Fundomo, with participation from Arm, at a $3.3 billion valuation. The round, led by 25-year-old founder James Dacombe, more than triples the company’s previous valuation of over $1 billion following its $220 million raise in February.

Help wanted

A maintainer sarcastically apologizes for their free software not meeting expectations and refuses to change its license, while rejecting feedback about reach and stars. They dismiss an AI-generated pull request, restrict contributions to approved collaborators, and mock competing projects that rely heavily on AI-driven development and sponsors. They emphasize they work for love of software, not money, and criticize trendy “agentic” development practices.

Andy Pavlo joins ClickHouse to establish ClickHouse Labs

Andy Pavlo is joining ClickHouse to lead a new research group, ClickHouse Labs, which will collaborate closely with engineers to advance database technology, publish research, and optimize ClickHouse’s PostgreSQL-based managed service. The lab aims to achieve industry-research impact comparable to IBM Research and Microsoft Research.

SpaceX wins $1.6B Space Force contract for 18 Falcon 9 launches

The U.S. Space Force awarded SpaceX $1.6 billion for 18 Falcon 9 launches from Vandenberg through 2027, the largest publicly disclosed order under the Phase 3 Lane 1 program. The missions support the Space Based Sensing and Targeting portfolio, for which SpaceX is also building major satellite systems under separate contracts. The award underscores SpaceX’s dominant position as the only provider currently able to sustain the required monthly launch cadence.

The AI bubble is popping; we just don’t know it yet

Big tech’s Q2 earnings have fueled investor fears of an AI bubble, marked by wild stock swings, soaring capex, and shrinking free cash flow. Analysts advise IT teams not to go all-in on frontier AI products amid this volatile market.

Situational Awareness and the Impending Stock Market Volatility

The article argues that Situational Awareness’s dramatic rise and fall reflects an over-leveraged and over-concentrated AI stock market, not a divergence from a positive trend. It contends that Citadel’s block purchase of SA’s assets is not necessarily a bullish signal, as it could serve purposes like short covering or market-making. The piece concludes that SA’s trajectory exemplifies a reflexive feedback loop driven by excitement and leverage, predicting significant volatility ahead.

Amazon completes $50B investment in OpenAI

Amazon has completed a $50 billion investment in OpenAI. The deal represents a major financial commitment by Amazon to the artificial intelligence company.

help wanted

A sarcastic open-source maintainer rejects criticism and demands for license changes, feature additions, and popularity contests. They dismiss AI-generated pull requests, commercial sponsorships, and competing projects, emphasizing that they prioritize personal values over user expectations or reach.

MediaTek lines up $5B war chest for AI datacenter push

MediaTek is allocating $5 billion for an AI datacenter expansion, aiming to capture up to 20% of an $80 billion market with its upcoming ASIC chips. However, analysts remain skeptical about the company’s ability to achieve that goal.

The AI bubble is already popping; we just don’t know it yet

Big tech’s Q2 earnings suggest investors are deflating the AI bubble, as massive swings in major tech stocks and concerns over high capex, shrinking free cash flow, and chip availability rattle markets. The Register’s podcast discusses what these numbers mean for investors and offers advice for IT teams navigating the volatile market.

🖥️ Hardware & Infrastructure

Leaked doc: Microsoft plans Xbox 360 games on PC, Project Helix consoles, Xbox PCs, handhelds

Microsoft plans to bring Xbox 360 games to PC, enabling them to run on its next-gen Project Helix console, “Xbox PCs,” and handhelds, according to a leaked document. The rollout is expected gradually between 2027 and 2028, and the document also indicates Microsoft’s disc-to-digital program is close to launching.

Tensions over AI data centers in London, Europe’s biggest data hub, as housing, power, water strain

London, Europe’s largest data center hub, is facing growing tensions as AI data centers compete with homes for energy, space, and water. This strain on housing, power, and water resources is becoming an increasing concern for the city.

Japanese startups race to make defense drones as Tokyo cuts reliance on Chinese drones (91% market)

Japanese startups are entering defense drone production as Prime Minister Sanae Takaichi pushes to cut reliance on Chinese industrial drones, which hold 91% of Japan’s market. Firms like Eams Robotics and Terra Drone are partnering with foreign companies and localizing battery production to expand domestic manufacturing capacity.

Utah produced more power from solar than any other source in May, a new first

In May 2026, Utah produced more power from solar than any other source for the first time. Wind and solar generation are surging in the state’s power mix, with experts expecting the trend to continue despite cuts to federal clean energy incentives.

Wind and solar overtake fossil fuels in Germany for the first time

In 2025, wind and solar generated 44% of Germany’s electricity, surpassing fossil fuels (43%) for the first time, a milestone also seen across the EU. The shift reflects Germany’s Energiewende strategy, including its nuclear phaseout, though coal remains a significant challenge with a phaseout deadline of 2038 or earlier. Political resistance from the AfD and new gas plants as bridge technology add uncertainty, but Berlin remains committed to its climate-neutrality targets.

We Fixed UniFi’s Slow PPPoE Performance with PPPoE Half-Bridge

Most UniFi gateways, including the UDM Pro/SE/Max, UXG Pro, and EFG, lack PPPoE hardware acceleration, resulting in significantly reduced throughput—typically 1.2–2.4 Gbps—while only the UCG Fiber, with its MediaTek Filogic 880 SoC, can exceed 5 Gbps. Because PPPoE processing is single-threaded, even multi-core CPUs bottleneck on a single session, a hardware limitation that firmware updates cannot fully resolve.

CP/M-386 – CP/M for 386 protected mode, derived from CP/M‑68K

CP/M-386 is an early-stage port of CP/M for Intel 386 protected mode, derived from CP/M-68K. It features 32-bit protected mode, Ring-3 TPA, and VGA/serial console support, but currently lacks device drivers. The project aims for source compatibility and has achieved full BDOS parity with CP/M-68K 1.3 and CP/M 2.2.

ssh ssh.place

An SSH-based collaborative canvas at ssh.ssh.place lets anyone draw using only their SSH key, with no account or installation required. Users can place one colored cell every 15 seconds on a 200x60 grid; text is not allowed, and cooldowns are tied to the user’s SSH key.

Microsoft says 8 GB of RAM should be enough for anyone running Windows 11

Microsoft stated that 8 GB of RAM should be sufficient for anyone running Windows 11, adding new items to its list of improvements for the operating system. The announcement highlights the company’s ongoing efforts to optimize Windows 11 performance.

🛠️ Developer Tools & Open Source

Celebrating 45 Years of Kermit with the First New C-Kermit Release in 15 Years

The first new C-Kermit release in 15 years marks Kermit’s 45th anniversary. The update revives the long-dormant file transfer and terminal emulation protocol software, though specific features remain unannounced.

Use Task Runners for Common Coding Tasks

The article addresses the challenge of remembering different commands for common coding tasks across various repositories. It suggests using task runners, including bash scripts and make, to standardize workflows and apply muscle memory. The author provides an example of a simple bash script wrapper for tasks like building, testing, and formatting.

SearXNG in Rust

A Rust-based metadata search engine that queries DuckDuckGo, Brave, Startpage, and Yahoo concurrently, scrapes their HTML results, deduplicates by normalized URL, and ranks results using Reciprocal Rank Fusion (RRF). It can be used as a library or run as a server, exposing a JSON API for health checks and search queries. The project includes installation instructions, usage examples, and environment-variable configuration options.

Hoplite (YC S26) – Effortlessly deploy cloud coding agents

Hoplite is a new YC-backed platform that lets developers deploy coding agents in the cloud, porting over local sessions, memories, and MCP servers to streamline feature QA. The founders built a custom agent harness instead of using off-the-shelf tools, aiming to make cloud-based agents more performant and better suited for reviewing product output. It offers free credits with the code “HACKERNEWS” and currently focuses on improving onboarding and previews.

Product analytics (and evals) for agent sessions on your MCP

Armature (YC P26) is launching a product analytics tool for MCP servers that reconstructs full agent sessions, showing user intent, popular use cases, and frequent issues via a simple SDK integration. The tool, which evolved from a standalone testing product, enables developers to identify workflow problems and automatically generate evals to test fixes across models and harnesses before shipping.

Devtools must be open source

AI agents now make it easy to personalize software by downloading source code, making local changes, and automatically rebasing them on upstream updates via scheduled cron jobs. This reduces both startup and maintenance costs, dramatically improving the return on investment for custom software. The article demonstrates this with examples like editing the Shelley agent and a code-review tool for meat.dev.

Prevent cognitive debt by manually retyping LLM-generated code

The author avoids cognitive debt by manually retyping LLM-generated code into their editor instead of reviewing AI-generated pull requests. This slower process helps them build a mental model, catch errors, and maintain a spatial map of their codebase, accepting being only 2x faster rather than 10x.

The true power of regular expressions (2012)

The article challenges the common claim that “you can’t parse HTML with regex,” arguing that modern regex implementations are far more powerful than formal regular expressions. It explains regular grammars and their equivalence to traditional regex, then uses the Chomsky hierarchy to show how practical regex flavors can match languages beyond regular ones.

Bonsai: Janestreet’s UI Library

Bonsai is an OCaml UI library for building performant, reactive web applications, inspired by Elm and used widely at Jane Street. It separates state, incrementality, and rendering into composable primitives, enabling efficient incremental updates and flexible state management. Key benefits include shared frontend/backend types, component-specific styles, a templating language, and expressive automated testing.

Octane – React’s programming model, compiled

Octane is a compiled, performance-focused framework and successor to Inferno that eliminates the rules of hooks and dependency arrays by having the compiler track dependencies, allowing hooks behind conditions. It introduces a .tsrx file format but supports plain .tsx for incremental adoption, while not supporting React Server Components. The compiler optimizes rendering with direct DOM writes and keyed lists while preserving familiar React-like component patterns.

Rust project goals: Immobile types and guaranteed destructors

Rust plans to introduce new auto-traits—Move, Destruct, and Forget—that let types opt out of being moved or forgotten, encoding these capabilities as type properties rather than place properties like Pin does. This would simplify self-referential types and enable guaranteed destructors, supporting patterns such as safe scoped async spawn. The proposal includes compiler MVPs, RFCs, and real-world validation in the Linux kernel.

Isopolis – Isometric pixel map of SF

Isopolis is an isometric, pixelated map of San Francisco that uses mapping data from OpenStreetMap and neighborhood information from DataSF. The article displays the map, which appears to still be loading its content. It visualizes SF neighborhoods in a distinctive isometric pixel style.

Boris Cherny on Trying to Get Claude Code to Rewrite the Claude App

Anthropic’s Boris Cherny described using Claude Code to attempt a pixel-by-pixel rewrite of the Claude desktop app from Electron to Swift, a task that had been running for over two weeks. John Gruber criticizes the approach, arguing that the app’s core problem is its poor design, not just its non-native framework, so a direct port would still be a bad app. Gruber also notes the current Mac app remains sluggish, taking 30 seconds to launch.

Why Book Corners won’t sync contributions back to OpenStreetMap

Book Corners cannot easily sync user-submitted bookcase data back to OpenStreetMap because contributing data requires compliance with OSM’s import and automated-edits guidelines, not just an API call. This involves publishing an import plan, community consultation, addressing licensing questions, and maintaining ongoing accountability. The author acknowledges these safeguards protect OSM’s data quality but impose a significant operational burden on the small project.

Celebrating 45 Years of Kermit with the First New C-Kermit Release in 15 Years

Marking 45 years of Kermit, the first new C-Kermit release in 15 years has been announced. This release continues the legacy of the long-standing file-transfer protocol.

Projects I Wouldn’t Use Python For in 2026

A long-time Python advocate now argues it’s not always the best choice, presenting a 2026 talk at a Christchurch Python NZ meetup. The talk covers his history with Python, why he avoids it for certain projects, and the alternatives he uses, while still highlighting areas where Python remains strong.

bonsai: A library for building dynamic webapps, using Js_of_ocaml

Bonsai is an OCaml UI library for building performant, reactive web applications, widely used at Jane Street. It separates state, incrementality, and rendering into composable primitives, with components as purely functional state machines that are incrementally rendered. The library also supports shared types between backend and frontend, a templating language, and expressive automated testing.

Rust Project Goals: Immobile types and guaranteed destructors

Rust plans to introduce new auto-traits—Move, Forget, and Destruct—to make type capabilities explicit, letting types opt out of being moved or forgotten. This would simplify self-referential types and enable guaranteed destructors for patterns like safe scoped async spawn. The approach follows the precedent of the Sized hierarchy and will be validated through compiler MVPs, RFCs, and real-world testing in the Linux kernel.

This Month in KDE Linux: July 2026

KDE Linux moved its monthly updates to official KDE infrastructure and integrated a new full-stack QA test system that already caught two regressions. It also added kernel-hardening tweaks, performance improvements, and a tool for downloading Flathub packages to handle Java and DOS apps. Various bugs were fixed and documentation was updated.

KC3 v0.1.17 released

KC3 v0.1.17 has been released, bringing fixes for pthread operations, hash table iterators, JSON parsing, and 5x faster incremental compilation. It also adds new features including log hooks, bi-directional Facts synchronization over TLS, HTTP/HTTPS client methods, and RUsage process statistics.

Retries don’t fix eventual consistency

Retries and dead-letter queues are the wrong response to eventual consistency, because missing data is often just normal asynchronous ordering rather than a failure. Instead of treating absent prerequisites as errors, services should store incoming events and process them only once all required data has arrived. This avoids unnecessary replay infrastructure and solves the actual consistency problem.

Gleam for Python Programmers

This article serves as a tutorial guide for Python programmers learning the Gleam programming language, progressing from fundamental types and functions to advanced concepts like pattern matching, error handling, concurrency, and state machines. It features hands-on projects such as building a key-value store, a virtual machine, and a web API, along with appendices for further reference.

Practical Memory Safety

Fil-C, proposed as a memory-safe C target for Zig, provides only a limited form of safety on a spectrum, not the absolute safety some expect. Unlike Rust’s stricter guarantees, Fil-C permits overwriting memory within an allocation, which can lead to security issues. The author argues that practical memory safety requires intuitive rules without hidden footguns, a standard Fil-C does not fully meet.

As Larry Ellison bets the farm, Oracle says it loves AI-written code, just not in OpenJDK

Oracle supports using AI-generated code for debugging and reviewing, but prohibits submitting AI-written output to OpenJDK. The stance aligns with CEO Larry Ellison’s aggressive AI push, yet sets clear limits on its use in core Java development.

🌍 Tech Policy & Culture

White House: voluntary advanced AI evaluation framework deadline met, but no details (Axios)

The White House met its deadline to establish a voluntary framework for evaluating advanced AI models, but it has not disclosed the framework’s contents, participants, or when companies will use it. The framework is meant to guide AI developers on government engagement, covering confidentiality, cybersecurity, and early access, though model thresholds and other details remain classified. A staff-level meeting with companies to review the framework is scheduled for Tuesday.

Trump admin invites OpenAI, Google, Anthropic staffers to White House Tuesday to review AI oversight

The Trump administration has invited staffers from OpenAI, Google, and Anthropic to the White House on Tuesday to review a completed AI oversight framework. The framework establishes a voluntary process for AI labs to submit their models to the government before releasing them to partners and the public. The meeting comes just after the initial Aug. 1 deadline.

EU AI Act enforcement begins: pre-release AI model checks, market restrictions, fines for providers

The EU’s AI Act enforcement powers took effect, allowing the Commission to inspect general-purpose AI models before release, restrict market access, and fine providers up to €15 million or 3% of annual turnover. The rules apply to any GPAI provider in the EU regardless of location, affecting U.S. firms like OpenAI, Anthropic, and Google. Non-EU providers must also appoint an EU-based authorized representative.

Amodei worries staff join Anthropic for money, not mission, amid AI talent war

Competition for elite AI talent is intensifying, with researchers frequently jumping between top labs despite enormous pay and resources. Lilian Weng recently rejoined OpenAI after leaving Thinking Machines Lab, which has lost several co-founders, while Google, Meta, and others also face retention challenges. Anthropic CEO Dario Amodei has reportedly worried that new hires are motivated by money rather than mission.

Ofcom Chair Cheshire overseeing wide-ranging review to refit regulator for Big Tech, OSA enforcement

Ofcom Chair Ian Cheshire is leading a “wide-ranging” review to refit the regulator for new duties overseeing Big Tech and enforcing the Online Safety Act. He said the watchdog will need “more resources” to enforce online safety rules as it takes on a wider brief.

Hank Green criticized for using ChatGPT, says LLM usage ’not healthy for me or good for the world’

YouTuber Hank Green apologized after viewers noticed a suspicious phrase in a video, leading him to acknowledge using ChatGPT for research under pressure. He said the “pushback” line was meant for the episode’s guest, but admitted the criticism was fair and that he felt “mortified.” Green plans to reduce his video output, citing concerns about AI’s impact and his own unhealthy reliance on LLMs.

Hacker News with AI stories filtered out

hcker.news, a Hacker News reader, now offers an AI filter to remove AI-related stories from the front page. The filter uses three passes: automatic keyword/domain filtering, an agent review, and manual review of uncertain cases. It can also exclude GitHub repos that show signs of AI authorship, based on commit messages, contributors, and repository files.

Ten advances in mathematics and theoretical computer science

OpenAI announced free ChatGPT access for 100,000 scientists and mathematicians, and reported that an internal AI model, Astra, produced ten results resolving or advancing long-standing open problems in fields including geometry, coding theory, and complexity. These results include a disproof of the Erdős unit-distance conjecture, new sphere-packing bounds, and progress on group theory and lattice cryptography. Human researchers prepared the manuscripts and formalized the proofs in Lean, with the mathematical arguments generated by the AI.

Taylor Farms has rewritten its cyclospora statement four times in sixteen days

Taylor Farms repeatedly revised its Cyclospora outbreak statement over 16 days, with the original version acknowledging that FDA traceback pointed to a specific farm as the potential source. Subsequent versions removed that admission, falsely claimed an FDA apology, and introduced inconsistencies in recall details. The author argues the original statement stands as evidence of what the company knew, and later rewording cannot erase it.

The Future, Made in China

Kai-Fu Lee, CEO of AI company 01.AI, traces how U.S. and Chinese technology policy diverged, with Washington ceding AI strategy to Silicon Valley while Beijing pursued a state-driven industrial plan for global AI leadership. China’s recent advances—including BYD overtaking Tesla and Chinese AI models rivaling U.S. systems at lower cost—illustrate its growing dominance. Lee predicts this shift will tilt the geopolitical balance toward China.

Xkcd: Earth Temperature Timeline

The article, titled “Xkcd: Earth Temperature Timeline,” consists solely of humorous viewing instructions for xkcd.com. It advises using an outdated emulated browser setup, enabling ad blockers, disabling high-heat drying, switching from Airplane Mode to Boat Mode, and keeping caps lock on for security.

The fading American Dream: Visa uncertainty drives Indian tech workers back home

Indian tech workers are increasingly returning from the U.S. due to H-1B visa uncertainty and stricter rules, with returnees potentially outnumbering outbound migrants by 2026. However, India’s tech job market is also slowing, as active openings dropped 14% in June and AI disruption affects major firms, creating challenges for returnees.

EU enforces labeling AI generated content

The EU’s new AI transparency rules took effect, requiring companies to clearly label AI-generated content such as deepfakes using watermarks or markers, with fines for non-compliance. The rules aim to help Europeans distinguish real from fake content, while exempting personal use and artistic works. Tech firms like TikTok and Meta have already implemented their own labels.

MPs demand answers on Fujitsu’s inclusion in lucrative frameworks

MPs on the Business and Trade Select Committee have demanded the government explain Fujitsu’s inclusion in four public sector procurement frameworks worth over £54bn, despite its self-imposed moratorium on bidding for new government customers. Committee chair Liam Byrne asked for clarification by August 12 on whether Fujitsu can use the frameworks to win work from new public sector bodies and how its participation aligns with the moratorium.

Less Coffee, Better Sleep

Reducing coffee intake to one cup daily for a week dramatically improved sleep, mood, and energy, with the author falling asleep easily and waking refreshed. The initial day without caffeine caused extreme fatigue and headaches, revealing how heavily previous consumption had masked exhaustion. The author concluded that drinking coffee due to tiredness was a cycle, as coffee itself was causing the fatigue.

Don’t be a meat proxy

The article criticizes people who relay AI-generated responses verbatim in communication, calling them “meat proxies” who add no value. It urges readers to read, understand, and validate AI output before responding in their own words, especially in contexts like code review where mindlessly using AI shifts the work onto others.

More German than many Germans

A Turkish computer science student arrived in Hamburg expecting cold, rule-obsessed Germans based on stereotypes, but his internship experience—including a trusting flatmate and friendly colleagues—changed his view. He was offered a job, returned after finishing his studies, and felt welcomed back by familiar faces and even a clerk at the registration office.

OpenAI’s super PAC is funding AI-generated news site attacking industry critics

An anonymously operated news site called Acutus is staffed by AI-generated reporters and publishes mostly AI-written articles attacking AI industry critics. The site is linked to Targeted Victory, the firm at the center of OpenAI’s $125 million political operation, suggesting OpenAI’s super PAC may be funding it.

What are you doing this week?

The article invites readers to share their plans for the week, while also reassuring them that it is perfectly acceptable to do nothing at all.

The EU Commission has proposed letting users set privacy preferences once in their browser to replace intrusive cookie banners, but the tracking industry, led by Google, is lobbying member states and the European Parliament to block the plan. The article urges citizens to contact their representatives to support the proposal, while noting that other parts of the Digital Omnibus reform are problematic and not endorsed.

Going Backward

The article traces a hypothetical evolution of a Go function to iterate a slice in reverse, starting with a simple but memory-wasteful copy, then a closure-based iterator with O(1) memory, and finally a callback-based iterator. This progression demonstrates why the standard library’s slices.Backward uses an iter.Seq2 signature, highlighting trade-offs between simplicity, efficiency, and caller ergonomics.

Don’t Take the Black Pill (Text Adaptation)

The article explores the “black pill,” a nihilistic and hopeless mindset from incel culture that discourages young men from improving their dating lives. The author describes escaping this spiral by focusing on being genuinely attractive and caring to everyone, not just potential partners. He concludes that despite exposure to harmful ideas, young people can still choose positive ones.

NASA boss balks at billion-dollar estimate for recycled Moon rover

NASA’s administrator has pushed back against a billion-dollar cost estimate for repurposing two Mars testbeds into a single lunar rover. The plan to recycle the hardware into a Moon explorer may prove far from cheap, raising concerns about the project’s affordability.

Sci-fi authors Scalzi and Stross decry AI’s dystopian impact on their craft

Science fiction authors John Scalzi and Charles Stross have criticized large language models as theft, calling their creators “absolute scum” for using writers’ work without permission. They argue that AI systems put authors at risk of copyright infringement while making it harder for them to earn a living.

Meta straps on a ‘Kick Me’ sign by mistakenly taking down video by India’s prime minister

Meta mistakenly took down a video by India’s prime minister, drawing criticism. Other tech news includes DeepSeek doubling peak-hour prices for its new model, Australia raising fines for Big Tech, and Japan planning its next moon mission on a domestic rocket.