Ein kuratierter TLDR-Stream fuer deine Themen.

Without large amounts of compute, small labs can only compete through algorithmic efficiency. Magic's pretraining recipe is now more than 10 times more compute-efficient than that of leading open-weight base models. The startup believes that pretraining, agentic RL, and long-context are sufficient for building superhuman coding agents and automating AI research and development. This post discusses its pretraining and long-context work.

Anthropic's Threat Intelligence team has identified and disrupted several operations in the past several months where threat actors tried to use Claude for malicious activity. This report shares case studies from those operations and describes how malicious use of Claude has evolved since the company's previous threat reports. The report covers disruptions between December 2025 and August. None of the misuse cases involved the use of Claude Fable or Mythos-class models.

Anthropic's Economics team developed a model predicting AI's impact on the US economy through various scenarios ranging from moderate growth to extreme transformations. In scenarios of rapid AI-driven growth, knowledge workers face greater unemployment and wage stagnation, while non-knowledge sectors may see increased wages. The model highlights economic inequalities, with more gains potentially flowing to capital rather than labor.

Open Code Review is an AI-powered code review CLI tool. It originated as Alibaba Group's internal official AI code review assistant — over the past two years, it has served tens of thousands of developers and identified millions of code defects. It has been validated at massive scale. The agent can read full file contents, search the codebase, inspect other changed files for context, and produce deep reviews.

AI startups often accumulate sprawling prompts full of contradictions and ambiguity as teams continuously add instructions. Treating prompts like product and code, with modular sections for background, behavior, and output, can improve agent quality, reduce regressions, and lower costs.

Meta introduces Muse, a personal AI agent, powered by Muse Spark, to help users achieve goals by automating tasks like booking travel or sending emails. Muse operates securely on Muse Secure VM, ensuring data privacy with unique protections like the Sentinel agent overseeing actions. Muse will soon offer encrypted data with Muse Confidential VM and is available on iOS, Android, and muse.ai in the US.

SWE-2 pushes the Pareto frontier and achieves 50.0% on FrontierCode 1.1 Main1, while being 64% cheaper. It beats SWE-1.7 and Grok 4.6 on both score and cost, matches GPT-5.6 Sol and Fable 5/5.1 at a fraction of their price, and comes within a few points of GPT-6 Astra at a quarter of the cost.

Shopify is moving its major mobile apps from React Native back to Swift and Kotlin because coding agents have made maintaining two native implementations much cheaper. Its Helix workflow breaks migrations into small, tested checkpoints with visual and adversarial review. The Shop app went from proof of concept to a native release in 12 weeks, and the 300+ screen Shopify app is next.

Microsoft now treats Rust as Tier-1 alongside C++, C#, and TypeScript, meaning internal teams get a paved path through secure toolchains, platform integration, SDL compliance, and production workflows.

A growth engineering intern at Browserbase built an AI engine to automate its SEO and AEO workflow. 53% of search traffic came from just five posts, while only 17 of 77 posts had been cited by AI engines. BlogEO audits existing content, finds search opportunities, generates drafts, and tracks AI citations. Human approval remains in the loop. After 6 weeks, search impressions grew 5.8X, page-one queries grew 9.8X, and average blog position improved from 10.9 to 6.6.

Construction is labor-intensive. There has, unsurprisingly, long been interest in automating the process to reduce the amount of labor required. However, automation has only been successful in tasks that can be done off-site in factories and on-site tasks that are extremely factory-like. Almost nothing outside of these two categories has succeeded.

Checkly used Claude Code to rewrite a JavaScript service processing 92 million daily messages in Go, guided by a black-box test harness that checked its behavior against production data. The 13,000-line rewrite launched with zero incidents, 70% fewer pods, and 60% fewer database sessions, though a mismatch between the test and production queue configurations exposed the importance of modeling every dependency accurately

AI labs are releasing model updates at a faster pace, making it harder for businesses to evaluate which tools are worth adopting. Anthropic, Meta, Google, and OpenAI all launched updates in one week. The rapid rollout is creating “model fatigue” as teams spend more time comparing capabilities and performance.

Large language models are able to write almost perfectly correct code. It is pretty straightforward to get a model to generate code and then let that code be checked by hidden tests. Checking the sloppiness of this code often requires human intuition and time, and is an extremely difficult task in general. This post illustrates why this is by going through possible ways of measuring slop.

Companies are experimenting with cheaper open models on inference providers. There is a trend toward more investment in model routing to route simpler requests to cheaper models. The industry is seeing more knowledge-sharing sessions on how to use AI models cost-effectively. More and more companies are setting per-developer limits on AI usage.

Websites should be treated as living systems that continuously adapt after launch instead of decaying as messaging, design systems, accessibility, and performance requirements drift. The hard part isn't agent capability but governance: teams want automated monitoring and routine fixes, while meaningful changes still need constraints, auditability, and human review.

PI-Desktop is an early-preview desktop workspace for coding agents that keeps projects, sessions, reviews, and permissions in one local-first app. It supports multiple model providers, MCP, skills, subagents, and plugins, while routing privileged work through a reviewable permission layer.

Next-token prediction optimizes for the shortest path to an answer, not necessarily the best or safest one. Prompts and safeguards cannot fully solve this architectural limitation, and it's worth exploring alternatives such as world models.

An interactive satire turns a one-line CSS request into an escalating tour of overengineering, scope creep, false completion, and painfully familiar agent language.

Meta describes an AI “second brain” built from structured, auditable knowledge files rather than retrained model weights. The system separates facts from procedures, turns expert feedback into regression-tested updates, and reduces assessment work from days to minutes while keeping changes transparent and reversible.

A developer describes the strange shift from flow-state coding to orchestrating 20-plus agents, and why the software factory itself is becoming the product.

When the same prompt is sent to an AI model multiple times at zero temperature, variations in the results can occur due to the nondeterministic nature of floating-point arithmetic combined with server load dynamics. The model's output can differ based on how requests are batched and how the GPU processes the addition of numbers, leading to discrepancies despite identical settings.

Tesla recently made a post that simply wrote, 'Go for launch,' with a graphic showing a '10.1' date.

Hunter Biden is launching $LAPTOP on Coinbase's Base network on September 9. The token name references the 2020 laptop controversy. The tokenomics earmark 20% of the supply for investors who lost money on $TRUMP and include a conditional burn mechanism tied to 30 predetermined outcomes, including a Democrat winning the 2028 presidential election, Bitcoin reaching a new all-time high, and $LAPTOP's market cap surpassing Solana's. Up to 30% of total supply faces burns if all conditions are met, while missed targets redirect proportional token amounts to charity. The structure embeds outcome-based supply changes that place it closer to a prediction market instrument than a conventional memecoin.

AI excels at routine work but struggles to hold consumer attention, which depends on novelty and authenticity. As apps become cheaper to build, they will resemble disposable content, shifting success from retaining users to producing fresh hits and building lasting brands.

Free AI access can attract paying customers more efficiently than traditional ads by allowing users to experience a product's value firsthand. Lovable treats free offerings as customer acquisition costs, targeting a three-month payback period while implementing eligibility rules and fraud controls to mitigate abuse.

Product strategy is defined as much by what you refuse to build as by what you pursue. Make a few clear bets, name the sacrifices behind them, and use those choices to determine what earns a place on the roadmap.

AI makes weak product management faster and harder to spot. The fundamentals still win: understand the problem, focus on outcomes, and own the direction.

Getting ahead is about choosing work with leverage and compounding upside.

Writing things down gives AI agents a form of memory, helping them manage work without truly learning continuously. Humans still supply goals and moral judgment, and even the best systems cannot substitute for taking action.

AI is turning software into shareable, remixable content. The next app store may be the social feed itself.

AI is turning engineers from coders into orchestrators.

PLG companies often weaken growth when they start chasing enterprise revenue. Sales teams begin treating self-serve users as leads, while product teams lock valuable features behind higher tiers. Enterprise opportunities should instead emerge from product usage. Growing teams and expanding usage can signal when an account is ready for sales. Self-serve customers should still be able to discover and buy the product without unnecessary sales friction.

Product design with AI turns to slop unless constraints come first and the extra copy, lines, and icons agents love to add get removed. The approach favors iterating in a design tool like Figma rather than the codebase, building reusable component libraries, and validating results with preview deploys against real data. At Ref, the team also borrows interface patterns from existing products for context and stress-tests taste by critiquing drafts as a group until something holds up.

Project Oasis is Adobe's new web-based graphic design app, described as a modern workspace with brand-aware AI built into the workflow. A September 7 community post invites agency, freelance, and in-house designers into early access, with a private feedback channel and an NDA. Pricing, a full feature list, and a launch date remain undisclosed.

Apple has unveiled the iPhone Duo, its first foldable. Featuring a 7.6-inch inner display, a 5.4-inch outer screen, Touch ID, and an adaptable iOS 27 interface designed for split-screen and half-folded use, the wider device supports paired apps and tabletop video. Counterpoint estimates it could capture up to 25% of the foldable market by year-end, despite foldables accounting for less than 2% of global smartphone shipments.

Kombai is your AI design engineer for designing standout websites and product UIs, then shipping production-ready frontend in the same loop.

Ten recurring mistakes make brands look unprofessional, according to a new logo design guide, from skipping brand strategy to cramming too many ideas into one mark. Other flagged missteps include trendy typefaces, unnecessary effects, overly detailed forms, generic stock symbols, and designing only in color instead of testing black-and-white first.

John Ternus became Apple's CEO on September 1, succeeding Tim Cook, bringing hardware expertise in reliability, durability, and repairability to a company facing an uneven AI rollout.

OpenAI's ChatGPT Images 2.5 generates images up to 50% faster, with more natural lighting, richer textures, and steadier detail preservation across edits and long conversations. A new Sketch tool lets users draw a rough reference for the model to build on, while the Flare and Sunburst variants weigh speed against fidelity. The model, rolling out now in ChatGPT, ChatGPT Work, and Codex plus API access, also adds guardrails, C2PA metadata, and watermarking to flag AI-generated content.

Shopify is acquiring Tailwind Labs, bringing the team behind Tailwind CSS, installed more than 110 million times a week by firms like OpenAI and Cloudflare, in-house. The framework stays open source and MIT-licensed, with Tailwind's creator Adam Wathan citing a stable long-term home for millions of dependent developers as the goal. The deal follows layoffs of three of four engineers in January after revenue fell 80%, and leaves Tailwind Labs closing sign-ups for commercial products like Tailwind Plus.

Code discovered in the latest iOS 27 beta suggests Apple may bring advanced photography features to the Camera app, including exposure histograms, waveform monitors, focus peaking, zebra highlight warnings, and manual focus controls. The update could also add burst-style self-timer shots and image processing tailored for the rumored variable-aperture camera on the iPhone 18 Pro. If released, these features would significantly narrow the gap between Apple's built-in Camera app and professional third-party camera apps.

Apple's marketing for the new $1,999 foldable iPhone Duo has drawn criticism after promotional images appeared to use unrealistically long fingers to make the device seem easier to hold one-handed. The exaggerated visuals sparked jokes and accusations of misleading marketing, adding to the debate over the Duo's ergonomics alongside its price and unconventional design.

Apple has unveiled the iPhone 18 Pro and Pro Max with the new A20 Pro chip, a smaller Dynamic Island, improved battery life, a 48MP variable-aperture camera with expanded manual controls, and AI-focused enhancements powered by a 32-core Neural Engine. Other highlights include faster graphics, better thermal management, support for Apple Reference Image authenticity metadata, up to 2TB of storage, and new accessories, with preorders opening September 12 and devices launching September 18.

A series of explorable worlds, each grown from a single picture. Click anything in the frame, choose what happens to it, and AI generates the next scene and a short film of the moment in between.

Archyl maps every system, service, and dependency into an interactive C4 model, then keeps it in sync on every push.

German software company SAP has released an update that patches 20 security vulnerabilities. Of particular note is a maximum-severity buffer overflow vulnerability in the External Passport Protocol (EPP) processing library, which could allow an unprivileged attacker to run arbitrary commands as an administrator. Another missing authentication vulnerability in the SAP NetWeaver Message Server could allow unauthenticated attackers to execute malicious payloads and arbitrary commands across the full SAP system cluster.

There is a malware campaign targeting F5 BIG-IP APM devices that exploits a critical RCE vulnerability and injects a fileless web shell. A distinct installer deploys the payload by infecting the Apache executable and modifying SELinux configurations. ESET researchers identified the malware as PoisonedRefresh.

NVIDIA has announced CUDA Rust and introduced two specialized compilation tracks that allow developers to author and optimize GPU compute kernels directly in Rust. This release brings Rust's strong compile-time memory safety guarantees and modern tooling ecosystem to GPU-accelerated workloads. Developers implementing embedded vision, intensive sensor fusion, and real-time state estimation on mobile platforms can now exploit GPU compute power while minimizing memory-related crashes and runtime concurrency bugs.

Security researchers gained access to an Elasticsearch cluster containing an Advanced Passenger Information System (APIS) database with records belonging to 220M passengers and flight crew members. The researchers received a 401 when trying to access the cluster via the public internet but could reach the database, which accepted default credentials via a cloud-based access path. The data includes passengers' and crew members' names, dates of birth, sex, nationalities, passport or travel document numbers, document expiration dates, and issuing countries.

Gamers Nexus published a two-hour investigation, done with security researchers MrBruh and uturn using packet capture and firmware analysis, alleging LG's smart TVs constantly log and upload user data even offline or in standby, scan local Wi-Fi networks for other devices, and record and store ambient audio through the TV's microphone for later retrieval, even when the TV appears off. LG called the claims untrue, saying its TVs only process voice data when the remote's voice button is pressed or its "Hi LG" wake word is detected after a user manually enables far-field voice recognition, with wake-word audio processed entirely on-device and deleted immediately if no match is found.

Calif has demonstrated a zero-click worm that takes over a WeChat account via an unanswered VoIP call, then uses that account to call the victim's contacts. This exploit relies on a memory corruption bug in WeChat's VoIP stack and requires the attacker to be on the victim's friend list, which is a weak barrier since one compromised friend can lead to further exploitation. Tencent has mitigated this issue in server-side updates and in client versions Android 8.0.77 and iOS 8.0.76, so users should enforce these updates and reconsider trusted-contact privileges in messaging apps.

New CEO John Ternus led his first keynote, headlined by the iPhone Duo, Apple's first foldable, a passport-style device with a 5.4-inch outer display and a 7.6-inch inner display that automatically reflows apps based on how the phone is folded, starting at $1,999. Apple also unveiled the iPhone 18 Pro and Pro Max ($1,199/$1,299, up $100 from last year) with new A20 Pro processors, a variable mechanical aperture camera, and 24-30 hour battery life, alongside the Apple Watch Series 12 and Ultra 4, which add a 5-second heart rate sensor Apple claims is more accurate than a chest strap.

OpenAI Korea GM Harrison Kim revealed the company has made real progress with Samsung Electronics on joint production and research for its next-generation chips, expanding a relationship that so far has been mostly about memory supply and enterprise software. OpenAI's current inference ASIC, Jalapeño, was designed with Broadcom and built at TSMC in under 18 months and is already in mass production there, so any Samsung involvement more plausibly targets its successor, which is reportedly approaching tapeout.

XPENG commissioned its humanoid robot production lines, and its IRON robot autonomously walked off them after completing assembly, a step the company frames as moving from R&D prototypes to real line manufacturing.

TSMC posted NT$514.8 billion ($16.1 billion) in August revenue, up 53.3% year-over-year and 10.1% from July, hitting a fresh monthly record as AI-driven chip demand keeps outpacing supply. It came alongside a separate announcement with Dutch equipment maker ASML: TSMC plans to bring ASML's High-NA EUV lithography into large-scale manufacturing starting in 2030 for its most advanced nodes, as AI accelerators push toward increasingly complex transistor architectures.

Besxar, a $9.35 million seed-funded startup, flew its first mission on July 5 aboard a SpaceX Falcon 9 alongside a Starlink batch, carrying its own semiconductor wafers plus test wafers from the University of Virginia and UT Austin. The company's pitch is simple: Earth-based fabs spend billions on pumps to create the vacuum semiconductors need, but in space vacuum is free, so Besxar wants to manufacture higher-quality, lower-cost wafers in orbit and return them to Earth to sell.

Anthem is a long-range precision strike weapon aimed at a gap Covenant argues current Western arsenals lack.

Google plans to invest at least $15 billion in Finnish AI infrastructure through 2028, its largest single investment in Europe, building on over 15 years of presence there. Alongside the buildout, Google signed a 22-year power agreement with Finnish utility Fortum and said it will explore business models that could support new nuclear reactors at Fortum's Loviisa site, tying the compute expansion directly to a long-term power source.

AMD and Embedded LLM teams took MiniMax M3's day-0 vLLM implementation from 109.1 to 342.4 output tokens/s/GPU on AMD's Instinct MI355X, a 3.14x gain, by repeatedly asking five questions as the bottleneck kept shifting: what local shape reached each rank after tensor-parallel sharding, what work repeats every layer, which bytes actually move, whether a claimed fast path actually executed, and which queue grows once kernels stop being the limiter. Big wins came from fusing the model's shared expert into the same grouped GEMMs as routed experts (up to 30% faster at low concurrency), and reusing sparse-attention block selections across adjacent layers instead of recomputing per layer.

White-hat hackers drained ~$320 million from Liquid Network's federation wallet over the weekend.

AI has shifted left from code repositories to coding agents on developer endpoints with broad privileges, where AppSec has limited control. AI-driven discovery is also flooding defenders with vulnerabilities. The better approach is to eliminate entire vulnerability classes rather than fix tickets one by one, while rethinking human-in-the-loop oversight when agents can both write and remediate code.

Oracle's cloud infrastructure revenue grew 121% year over year as demand for AI compute continued to surge. The company spent $28.5 billion on capital expenditures during the quarter and maintained plans for $90–95 billion in spending this fiscal year

Accenture and Google Cloud launched a new Gemini Enterprise group, including a 1,000-person FDE workforce, to scale agentic AI adoption. The group will help enterprises move from AI pilots to production with industry-specific solutions and implementation support. YouTube saw 11% higher customer sentiment and 37% lower handle time using Gemini Enterprise.

Claude for Microsoft 365 integrates Claude into Excel, PowerPoint, Word, and Outlook with shared context across apps. Claude for Excel, PowerPoint, and Word is now generally available on all paid Claude plans, while Claude for Outlook is in beta. The offering emphasizes reviewable edits, use of customer templates, and deployment through existing cloud and compliance frameworks.