Daily Tech News - 2026-09-20

Tech’s center of gravity is shifting from bigger clouds and bigger models toward tighter control, faster execution, and infrastructure developers can actually trust.

AI tooling dominated the day. Google’s Open Agentic Orchestrator, AX, promises to run sandboxed agent tasks at massive scale, while experiments with “software factory” workflows and parallel IBM Bob agents show developers turning autonomous coding into a repeatable production process. Yet trust remains the defining constraint: Mira’s VPC deployment keeps code review inside controlled infrastructure, and a critique of MCP questions whether today’s favored integration layer is sound at all. TypeSafe AI’s JEV classifier points toward specialized routing tools that claim dramatic speed and cost advantages over general-purpose LLMs, while GraphRAG practitioners are refining more sophisticated ways to combine graphs, semantic retrieval, and model reasoning.

That appetite for control extends beyond AI. One developer built a home server from spare parts after growing frustrated with cloud storage pressure, while another deployed GoatCounter on NixOS for self-hosted, bot-filtered analytics. New writing on sandboxing—and research into adversarial inputs for fast hash functions—underscores that local ownership still demands serious security engineering.

Performance and preservation provided another strong thread. NumPy in the browser is closing a longstanding acceleration gap, w64devkit reported a productive year of Windows tooling improvements, and Linux 7.3 received important last-minute x86 fixes. Meanwhile, reverse-engineering communities reached striking milestones: Resident Evil 4 achieved a byte-identical GameCube decompilation, and the Ogre Battle 64 recompilation project passed 99%.

Elsewhere, the jj ecosystem, Lambda MicroEgg’s binder-aware e-graphs, CBAM implementations, and preparations for the Unix Year 2038 problem all reinforced the same lesson: durable computing advances through careful tools, explicit constraints, and relentless attention to details everyone else overlooks.

Featured Articles

Beyond jj: config & tools ecosystem

16 Sep 2026 Beyond jj : config & tools ecosystem This post was originally given as a talk at JJ Con 2026. The slides are also available. Hello! Welcome to “beyond jj”, where we’re going to take a look...

• Keywords: jj commands, jj config, configuring jj, command jj, jj tools, jj command, development jj, create jj, official jj, jj create
• Source: andre.arko.net

Custom home server built from spare parts

Google nagged me into this, along with every other enshittified cloud provider dealer. For about a year, my phone warned me that the account was nearly full, so I migrated the picture backups to a Nex...

• Keywords: picture backups, backups nextcloud, backups, backup, automatic backup, like backups, keeps backups, backups downloads, backups turned, gpus
• Source: asmat.ca

The last mile of a long road: faster NumPy in the browser

The last mile of a long road: faster NumPy in the browser For a long time, running NumPy in the browser meant running it without an accelerated BLAS. Matrix multiplications fell back to plain loops (p...

• Keywords: faster openblas, speedup openblas, speedups openblas, numpy builds, numpy benefits, openblas optimized, openblas numpy, numpy browser, numpy openblas, faster numpy
• Source: notebook.link

Software sandboxing: The basics (2025)

Diving into the territory of software sandboxing is diving into mostly uncharted territory. The necessary pieces to implement good sandboxing in your software are scattered all-around and the pioneers...

• Keywords: sandboxing security, practical sandboxing, basics sandboxing, sandboxing concerns, sandboxing techniques, definition sandboxing, general sandboxing, software sandboxing, employing sandboxes, builtin sandboxing
• Source: blog.emilua.org

Run Two IBM Bob Agents at Once with Git Worktrees and Podman

I want my coding agents to run unattended. But I have to admit that I have trust issues also. And this is why the initial question for this article came up: could I containerize and secure Bob Shell 2...

• Keywords: unattended agent, trust workflow, trust unattended, container git, containers git, run unattended, agent bin, unattended work, enforcing podman, podman exec
• Source: the-main-thread.com

What's been going on in w64devkit the past year

nullprogram.com/blog/2026/09/20/ The past year has been exciting for w64devkit, which like any software distribution is never complete. Peter0x44 joined as co-maintainer, and has pushed the project in...

• Keywords: w64devkit releases, w64devkit patches, dlls w64devkit, w64devkit produce, unique w64devkit, w64devkit older, exciting w64devkit, w64devkit bin, runtimes w64devkit, w64devkit added
• Source: nullprogram.com

Adversarial examples for fast hash functions

Contents Hash functions map data of arbitrary length to fixed-size values. The goal is to ensure that distinct inputs map to distinct outputs, except with very small probability over the randomness of...

• Keywords: hashing useful, quality hashing, hashing needs, quality hash, hash robust, throughput hashes, hashes fasthash, hash better, best hashes, secure hashes
• Source: thomasahle.com

Bot-free self-hosted analytics with GoatCounter on NixOS

Bot-free self-hosted analytics with GoatCounter on NixOS Vincent Bernat 21-minute read Also available in Filed under Attachment: In 2016, I removed Google Analytics from this blog to avoid being compl...

• Keywords: analytics goatcounter, proxy goatcounter, goatcounter proxy, hosted analytics, goatcounter internet, goatcounter privacy, goatcounter web, goatcounter server, website goatcounter, site goatcounter
• Source: vincent.bernat.ch

JEV Classifier: 5.43x as Fast as Haiku, 96% Lower Cost

JEV Classifier: 5.43x as Fast as Haiku, 96% Lower Cost Last Updated: September 18, 2026 An Auto Router pays for classification before the selected model can answer. In our benchmark, TypeSafe JEV clas...

• Keywords: classifier latency, classification speed, classifier jev, jev benchmark, jev classifier, jev classification, benchmarks jev, routing_decision classifier_cost, latency tier, classifier cost
• Source: docs.litellm.ai

Jev and LLMs: who does what?

After two years in stealth, Diogo Almeida, one of the researchers behind ChatGPT, launched Jev at TypeSafe AI. The company advertises roughly 200 times faster and 400 times cheaper decisions than LLMs...

• Keywords: evaluating jev, jev evaluates, uses jev, compare jev, ai jev, jev classification, experimental jev, jev use, jev engineering, slow jev
• Source: generativeprogrammer.com

Google's Open Agentic Orchestrator

Declare an agentic task. AX runs it at scale. AX sandboxes your task, wires up its workspace, fences its network, and helps you run billions of them per cluster. Either use a single task per agent, or...

• Keywords: tasks ax, ax tasks, task ax, ax task, scheduling ax, workspaces golang, ax agents, running ax, task yaml, task agent
• Source: agentexecutor.io

GraphRAG: A Practitioner's Guide to 6 Advanced Architectural Patterns

GraphRAG: A Practitioner's Guide to 6 Advanced Architectural Patterns Beyond basic graph retrieval: six production-oriented architectures for combining semantic search, knowledge graphs, and LLM reaso...

• Keywords: semantic queries, semantic search, entities retrieval, relational rag, retrieval production, search knowledge, retrieval systems, reasoning retrieval, query unstructured, queries knowledge
• Source: towardsdatascience.com

Unix Year 2038 problem and the art of underestimating

Runtime Arguments Conversations about technology between two friends who disagree on plenty, and agree on plenty more. Runtime Arguments 37: Unix Year 2038 problem and the art of underestimating Use L...

• Keywords: hitting 2038, running 2038, 2038 fix, 2038 thing, 2038 think, unix historically, unix year, bit timestamps, 2036 problem, time sector
• Source: buzzsprout.com

Resident Evil 4 (GameCube) – complete byte-identical decompilation to C/C++

A complete, byte-identical decompilation of Resident Evil 4 for the Nintendo GameCube: the G4BE08 debug build (the "Nov 25 2004" prototype, both discs), whose Bio4.sym files name every function. Build...

• Keywords: cp re4_debug_disc1, re4_debug_disc2, re4_debug_disc1, decomp toolkit, re4_debug_disc1 iso, codewarrior builds, tools decomp, re4_debug_disc2 gcm, iso re4_debug_disc2, compiled prodg
• Source: github.com

How to Generate Code with ChatGPT

ChatGPT is not limited to answering questions or writing content. You can also use it to generate code, explain programming concepts, debug errors, write SQL queries, and learn new programming languag...

• Keywords: chatgpt programming, programming chatgpt, development chatgpt, chatgpt programmers, programmers chatgpt, program chatgpt, code chatgpt, chatgpt coding, chatgpt create, chatgpt generate
• Source: javaguides.net

Why MCP Was Always a Bad Idea

Why MCP Was Always a Bad Idea Why MCP Was Always a Bad Idea Recently I went to an all-day event centered around the latest and greatest in the MCP world. While all of the presenters were awesome and s...

• Keywords: mcp industrial, service mcp, idea mcp, mcp world, use mcp, mcp bad, replace mcp, think mcp, today mcp, mcp continued
• Source: maharship.com

grimjim/Llama-Nephilim-Metamorphosis-v1-8B — public verification state: verified

grimjim/Llama-Nephilim-Metamorphosis-v1-8B grimjim/Llama-Nephilim-Metamorphosis-v1-8B is an open-weights language model for text generation under the llama3.1 license (15.0 GB), released Oct 14, 2024....

• Keywords: model grimjim, llama3 license, license llama3, 24gb grimjim, 80gb grimjim, 8b grimjim, model text, generation grimjim, download grimjim, grimjim released
• Source: huggingbay.xyz

Lambda MicroEgg

Lambda MicroEgg It’s an egraph that supports well-scoped alpha aware binders. Everything old is new again. I made a tool that attaches my lifting e-graph ideas arxiv youtube to an s-expression based f...

• Keywords: microegg egraph, binders sum, patterns addition, lambda microegg, patterns binders, sexp rewrite, sum rewrite, pattern improved, substitution routine, patterns extending
• Source: philipzucker.com

Trying the Software factory pattern.

Trying the Software factory pattern. One of the interesting challenges of the AI ecosystem in 2026 is that new, effective patterns emerge faster than I can adopt them. I’ll find a handful, get back to...

• Keywords: software factories, software factory, development heavily, factories agentic, agent driven, ai ecosystem, agents iterate, patterns emerge, adopt software, agents
• Source: lethain.com

A VPC deployment option keeps AI code review inside controlled infrastructure

AI code review is harder to trust when you can’t control where the code goes. Mira is a self-hosted AI code reviewer for teams that want to review pull requests with their own infrastructure and LLM s...

• Keywords: code reviewer, code review, reviewer teams, reviewer project, review pull, review changes, auto reviews, github, gives reviewer, custom review
• Source: x.com

CBAM Paper Walkthrough: The Double-Attention Mechanism

CBAM Paper Walkthrough: The Double-Attention Mechanism Understanding and implementing CBAM (Convolutional Block Attention Module) from scratch with PyTorch Introduction In this article, I am going to...

• Keywords: cbam attention, cbam convolutional, attention modules, cbam improves, cbam enables, uses cbam, attention module, cbam architecture, cbam used, cbam implement
• Source: towardsdatascience.com

Ogre Battle 64 Recompiled Project at 99.05%

Static recompilation of the N64 game Ogre Battle 64: Person of Lordly Caliber (USA, Rev A) to a native PC executable, using the N64Recomp toolchain. This repository contains no copyrighted game data....

• Keywords: recompilation n64, n64recomp build, recompile n64recomp, n64recomp tools, n64recomp modifications, n64recomp toolchain, patch n64recomp, 64 recomp, using n64recomp, tools n64recomp
• Source: github.com

Pirate Face Rescues LLM Models from Deletion

Turn AI into torrents that live forever. Open models - LLMs, image, audio, datasets - as 🧲 magnet links that can never be taken down. No single owner or point of failure. Trending Models Trending Apac...

• Keywords: ai torrents, ai pirate, torrent model, torrent permanence, model torrent, torrent guaranteed, torrents, torrents live, link torrent, create pirate
• Source: pirateface.co

Important x86 Fixes Merged Ahead Of Linux 7.3-rc4

Important x86 Fixes Merged Ahead Of Linux 7.3-rc4 Ahead of the Linux 7.3-rc4 kernel release due out in the coming hours, an important batch of x86/x86_64 fixes have been merged to address a few outsta...

• Keywords: affected microcode, microcode versions, microcode change, microcode revision, microcode update, microcode version, x86 fixes, problematic microcode, runtime microcode, 0x1000423 microcode
• Source: phoronix.com

Pi 0.86.1

Meta Muse provider — Sign in with Meta using /login meta or use META_API_KEY to access Muse Spark models. See Meta (Muse subscription). Added Added Meta (Muse subscription) login via /login meta with...

• Keywords: meta muse, muse provider, access muse, muse subscription, muse spark, muse, meta_api_key support, containers wsl, meta_api_key, models meta
• Source: pi.dev

Mathematical Billiards (2024)

STRUCTURES Blog > Posts > Mathematical Billiards Mathematical Billiards Do you enjoy a game of pool or billiards? I certainly do; I find it very satisfying to pocket a ball with a complicated shot fol...

• Keywords: billiards models, billiards study, mathematical billiards, idealized billiards, physical billiards, billiards mathematical, billiards consider, billiards wikipedia, billiards games, table billiards
• Source: structures.uni-heidelberg.de

More dirty coding tricks from game developers (2015)

In this timeless feature from the March 2010 issue of Game Developer Magazine, we revisit the mighty kludges and well-meaning hacks that are occasionally required to get our games into the hands of ea...

• Keywords: feature hacks, coding hacks, game disaster, release games, bug game, terrible hacks, hacks save, issue game, save games, hack player
• Source: web.archive.org

Comparing reflection capabilities of C++, Zig and C3

Comparing reflection capabilities of C++, Zig and C3 Comparing reflection capabilities of C++, Zig and C3 Reflection lets a program inspect and manipulate its own structure at runtime or compile time....

• Keywords: reflection c3, c3 reflection, c3 comparing, c3 uses, c3 rely, development c3, c3 versions, official c3, zig c3, c3 macros
• Source: nyr24.github.io

If AI coding is lowering your code quality, you're not managing quality right

One common take on the coding agents that I see goes something like this: “Sure, AI helps you output more code, but won’t the quality suffer?” It certainly will if you just blindly merge the PRs and s...

• Keywords: coding agents, bugs control, keeping bugs, instruct agent, defensive layers, managing quality, code quality, maintained developers, code production, agents make
• Source: i-kh.net

Deterministic Core, Non-Deterministic Shell

Fourteen years ago, Gary Bernhardt coined the term Functional Core, Imperative Shell . Like most good ideas in computing it was not entirely new, but his conception had great clarity, and it forms an...

• Keywords: functional core, core imperative, imperative shell, shell core, shell architecture, functional io, core purely, deterministic core, behaves functional, purely functional
• Source: outdata.net

StepFun details Step 5 Preview’s long-horizon software performance

Step 5 Preview works across software environments and sustains execution over long horizons. Its capabilities extend from software engineering and web applications to 3D workflows and programmable har...

• Keywords: step preview, execution feedback, preview, 3d workflows, gpu kernel, preview works, preview keeps, workflows programmable, uses execution, gpu
• Source: x.com

TeksEdge runs DeepSeek V4 Flash and Qwen3.8 Flash-Next on one DGX Spark

📣 New Inferenceing Engine Alert! One DGX Spark is now running DeepSeek V4 Flash + Qwen3.8 Flash-Next at 262K context with ~1,000 tps PP. A new engine called Athena recently dropped for Nvidia GB10 sys...

• Keywords: athena caches, dgx spark, spark deepseek, 128gb dgx, tps 256k, deepseek v4, running deepseek, performance athena, flash 8k, qwen3 flash
• Source: x.com

[email protected]

93e31fc: Add ai.ask.cors to let listed origins (or "*") call the generated Ask AI endpoint from another site. The route answers the browser’s preflight and names a listed origin on every response, the...

• Keywords: reasoning_effort openai, openrouter reasoning_effort, ai endpoint, gateway reasoning, reasoning_effort, route answers, effort openrouter, ai sdk, add ai, reasoning effort
• Source: useblume.dev

Joy & Curiosity #100

It’s the week of Jev! I’m really, really, really, really excited about it. I mean: really. It’s like someone blew up a confetti bomb in the world of LLMs and now you realize how grey everything looked...

• Keywords: jev predict, jev llm, uses jev, jev picking, jev really, think jev, jev, jev feels, excited jev, shell jev
• Source: registerspill.thorstenball.com

ModelScope confirms Qwen-Image-2.1 weights and code will be released

[email protected] is going open source! 🎨 ⌛Join the countdown modelscope.cn/models/Qwen/Qw… Weights + code. Download it, run it, build with it. 💻2:35 AM · Sep 20, 2026·62.3KViews...

• Keywords: modelscope2022qwen image, modelscope20223h, countdown modelscope, modelscope20223h count, modelscope modelscope20223h, modelscope2022qwen, modelscope ai2162, modelscope modelscope2022qwen, 3kviews1564519221 modelscope, modelscopefrom
• Source: x.com

Quarkdown: Turing-complete Markdown typesetting system

• About

• Demo

• Editor extensions

• Targets

• Comparison

• Getting started

• Mock document

• Contributing

• Sponsors

• Concept

• License Quarkdown is a modern Markdown-based typesetting system design...

• Keywords: quarkdown markdown, markdown combined, functions markdown, extension markdown, markdown based, markdown create, markdown, modern markdown, known markdown, markdown syntax

• Source: github.com

How Notion handles concurrent editing with CRDTs

Notion’s editor is commonly used in a collaborative setting, as teams of hundreds and thousands write and work together. But until 2025, Notion wasn’t really collaborative. To help people work more se...

• Keywords: editing notion, concurrent edits, notion editor, collaborative editor, editing concurrent, editing systems, concurrent edit, notion blocks, notion block, occurring editing
• Source: notion.com

Orchestrating Claude Code Agents: The Chief of Staff Pattern

Long-horizon AI coding work fails less because agents cannot write code and more because their context is ephemeral and their self-reports are unreliable. The fix is organizational rather than technic...

• Keywords: agent orchestration, agent runtime, agents write, coding agents, orchestration execution, agent sessions, failure agentic, agent operates, orchestration coordinator, scriptable coordinator
• Source: asyncdot.com

Is Udacity’s Agentic AI Nanodegree is the Best Agentic AI Resource in 2026? Review

Hello friends, Agentic AI is probably the biggest shift happening in AI right now. A year ago, most developers were building simple AI chatbots. Now? Companies want developers who can build:

• AI agen...

• Keywords: ai agent, ai trends, ai agents, ai nano, 2026 ai, ai engineering, ai development, build ai, agentic ai, ai industry

• Source: java67.com

OpenMote: ESP32 board for a WiiMote shell

Back this project to help bring it into existence. Funding ends on Oct 29, 2026 at 05:00 PM PDT. OpenMote is a programmable universal remote in a nostalgic form factor. The same remote runs your TV, y...

• Keywords: wii remote, openmote remote, remote openmote, wiimote, requires wiimote, shell wiimote, wiimote shell, openmote programmable, bluetooth remote, board wiimote
• Source: crowdsupply.com

Using non-breakable spaces in test method names

Using non-breakable spaces in test method names Yes. This article is about using non-breakable spaces to name tests. And the fact that it's awesome. And why you should use them too. public function te...

• Keywords: spaces tests, naming test, function test_a_user_can_add_a_product_to_a_wishlist, test methods, test_a_user_can_add_a_product_to_a_wishlist, testaddproducttowishlist kind, function testaddproducttowishlist, test method, spaces test, breakable spaces
• Source: mnapoli.fr

Andrew Chen says Jev could unlock ad-supported AI-native apps

Jev is going to change the prosumer/consumer AI landscape by unlocking a specific thing: Ad-supported + free AI native apps this is going to usher in a generation of new AI-native marketplaces, social...

• Keywords: ai costs, subsidize ai, free ai, ai native, consumer ai, ai, ai landscape, new ai, apps smarter, ad supported
• Source: x.com

Qwen-Image-2.1 in ComfyUI: Open-Weight Image Generation and Editing, Now with Transparency

A new open-source image editing model has finally arrived. Qwen-Image-2.1 is now supported natively in ComfyUI. Open weights, 7B, and it generates a real alpha channel. Qwen-Image 2.1 is the latest im...

• Keywords: qwen image, image editing, encode qwen, qwen, download qwen, channel qwen, image model, images 7b, input images, load qwen
• Source: blog.comfy.org

Samsung is expected to more than double output of its HBM4 and HBM4E DRAM

Samsung Electronics (005930) is expected to more than double output of its HBM4 family of high-bandwidth memory chips next year, including sixth-generation HBM4 and seventh-generation HBM4E, as the co...

• Keywords: glass carriers, samsung hbm, carriers glass, hbm dram, glass carrier, samsung preparing, material hbm, samsung increase, glass supports, hbm production
• Source: en.sedaily.com

Qwen-Image-2.1: Compact, efficient, and unified image creation

<a href="https://news.ycombinator.com/item?id=49775499">Comments&lt;/a>

• Keywords: news ycombinator, href, ycombinator com, ycombinator, href https, comments, https news, 49775499 comments, com item, news
• Source: qwen.ai