0:08
Cloudflare and AWS Embed x402 Agent Payments at the Edge
Cloudflare and AWS both rolled out the x402 stablecoin micropayment system on their edge networks within a fortnight, and the speed of that rollout is what caught my eye. The protocol, sitting under the Linux Foundation’s umbrella, essentially brings back the old HTTP 402 status code, but now it’s being used for tiny, sub‑cent transactions between agents and services. It’s a neat little twist on something that’s been around for ages, repurposed for a very modern use case.
What’s striking is how quickly the ecosystem is moving. Coinbase says they’ve already logged 169 million of these micro‑transactions in the first year alone, which suggests the network effects are kicking in faster than many expected. The payments are stablecoin‑based, so the value stays predictable, and the edge‑centric design means the latency is minimal—perfect for the kind of real‑time interactions developers have been craving.
The only big hiccup left is the back‑office side of things. Enterprises are still wrestling with how to fit these micro‑payments into their existing tax and invoicing frameworks, and that gap hasn’t been closed yet. It’s a reminder that while the tech can be slick, the surrounding processes often lag behind. Still, the groundwork is there, and the next steps will probably focus on tightening those compliance loops.
1:15
Google DeepMind's design leader used Claude Fable 5 to port Command and Conquer: Generals Zero Hour to iPhone and Mac
A DeepMind design lead just fed Claude Fable 5 a chunk of the original Command & Conquer: Generals Zero Hour code, and instead of wrapping it in an emulator, the model rewrote the engine to compile straight to Arm64. That means the game runs natively on iPhone and iPad, using the device’s own CPU cores rather than a translation layer.
The trick was in how the AI parsed the C++ build scripts, resolved platform‑specific calls, and regenerated the binaries with the right instruction set. It wasn’t a simple port; the AI handled the low‑level glue code that usually trips up human developers, turning a Windows‑centric build into a clean, modern iOS executable.
The result is a fully playable Zero Hour that feels like the original, but with the smoothness you expect from a native app. It’s a neat glimpse of how generative models can take legacy code and give it fresh hardware life without the usual hand‑wrangling.
2:07
How Tech Scammers Conned Four People Out of $673,000 in Three Days
USA Today reports on a Facebook post from a Washington state sheriff's office:
Four residents of Clallam County, a coastal region west of Seattle along northern Washington's peninsula, lost more than $673,000 in just three days, according to the Clallam County Sheriff's Office... The smallest amount lost was $3,500, which someone purchased in Apple gift cards for a scammer posing as an employee with Microsoft technical support, the sheriff's office wrote. Another person lost $50,000 after they clicked on a malicious email and unwittingly granted the scammers access to their financial accounts.
The local Peninsula Daily News reports another scam involved a 64-year-old resident who attempted to contact Coinbase after seeing their account displayed shown as closed:
"Believing they were speaking with a legitimate Coinbase representative, the victim was told there was fraudulent activity on the account and was instructed to download a 'rescue' application," the [sheriff's] release states. "The application allowed the scammer to remotely access the victim's phone." They then convinced the victim to transfer approximately $200,000 worth of cryptocurrency to what was described as a secure wallet. The funds were instead transferred to the scammer and could not be recovered...
In one scam, reported Monday, an 84-year-old Clallam County resident believed they had received an email from their daughter with a photo. After opening the email, a fake Microsoft security alert appeared on the computer directing the victim to call a support number, according to the release. "The victim was transferred to someone claiming to represent the Federal Trade Commission (FTC) and was falsely told they were under investigation in a child pornography and money laundering case," the release states. "The scammers instructed the victim not to contact local law enforcement and claimed local banks were also under investigation. The victim was told their bank accounts were in danger of being seized and was instructed to purchase gold to protect their assets." In three separate transactions, the victim purchased approximately $420,000 worth of gold and gave it to an unknown man waiting at the end of their driveway.
"Only after speaking with bank officials did the victim realize they had been defrauded," the release states.
USA Today offers this advice from the sheriff's press release. "These criminals are professional manipulators who prey on fear, trust and urgency. We encourage everyone to pause before sending money, purchasing gold or gift cards, or transferring cryptocurrency. A simple phone call to a trusted family member, your bank or local law enforcement can prevent a life-changing financial loss."Read more of this story at Slashdot.
Source:
4:22
Extract Fumes in Midcentury Style With Nixie Tubes and Military Surplus
Nobody wants to breathe solder fumes; that’s a given. For most of us, an industrial-looking fan-and-filter made in China and picked up cheap feels like more than enough to keep our lungs clear. Other people, people like [George Conneely], have more refined tastes. Why have a fume extractor when you can have a work of art?
This is one of those projects where the images really speak for themselves, because the whole point is to be beautiful. Sure, the wooden case is lovingly sculpted by a—wait, it’s 3D-printed!? Yes, with the right stain and care applying it, it seems Wood PLA can fool the eye, or at least the camera. Inside that PLA case there’s a custom PCB with an ATMega microcontroller and some MOSFETS to drive the Nixie tubes. The two digits represent the fan’s set RPM as a percentage of maximum, as is clearly labeled. Using a READY/NOT READY indicator pulled from a Panvia Tornado to show whether the fan has actually spun up to its set speed is an amazing touch.
The only problem with this build is that it is too nice. We’d almost rather see it on Don Draper’s desk than risk dirtying it on a lab bench. Evidently, [George] ascribes to the philosophy that one should surround oneself with beauty whenever possible. Your tastes may differ, but to many, nixie tubes certainly qualify– whether on a desk clock or in a car’s dashboard, there’s just something about that incandescent glow.
Thanks to [George] for the tip.
5:34
EY sacks staff for allegedly accessing Australian Prime Minister’s bank account
ASIA IN BRIEF The Down Under outpost of consultancy EY has fired two staff after they allegedly accessed details about bank accounts held by Australian Prime Minister Anthony Albanese. Local media report that the pair were employees of the artist formerly known as Ernst and Young, which used them to work on a contract for Australia’s Commonwealth Bank where they allegedly accessed details of the PM’s bank account. The Bank isn’t commenting on the matter, so isn’t saying how a pair of contractors were able to access such sensitive information. The incident means all four members of the so-called Big Four consultancies are in the doghouse down under. KPMG Australia’s chair recently left the organization after a whistleblower revealed it inappropriately used a client’s confidential data and Deloitte last year repaid consultancy fees after it submitted a report partly created with AI. PWC’s scandal is the worst of the lot as the company was asked to consult on tax policy and then used the knowledge gained during that engagement to advise tech companies on how to pay less tax. South Korea and Australia brace for the downside of the tech boom South Korea last week announced record monthly exports of over $100 billion, thanks largely to inflated memory prices meaning huge sales for Samsung and SK Hynix. News of that record was quickly followed by a warning from the Organisation for Economic Co-operation and Development, which in its annual Economic Survey [PDF] of South Korea warned “increasing dependence on semiconductor exports boosts growth and tax revenues but may also increase exposure to external shocks and cyclical volatility.” In Australia, meanwhile, the nation’s central bank revealed it has debated whether the local datacenter construction boom spurs inflation – which remains unpleasantly high. In minutes of its most recent board meeting, the Reserve Bank stated “Members noted that, while much investment in data centre requires imported components, it also requires some domestic inputs. They discussed the potential for continued strength in such activity to exacerbate capacity pressures and skills shortages in other parts of the economy.” India goes after Uber with national rideshare platform India’s government last week advanced its plan to create a national rideshare platform that charges no commission to drivers. Home Minister Amit Shah last week announced the service, called “Bharat Taxi,” will operate in seven cities by the end of July – then expand to 500 cities in the next 18 months to two years. India launched the service last year and said it would mean drivers earn more and reduce dependence on foreign platforms. The government statement announcing the service’s commencement alleges that rival private operators have dropped fares in “an attempt to block the entry of ‘Bharat Taxi’ so that they can again resort to arbitrary practices.” The minister resolved that Bharat Taxi will be around for the long haul. Amazon sued over Prime promises Also in Australia last week, the nation’s Competition & Consumer Commission launched a lawsuit in which it claimed that Amazon may have breached consumer protection laws by using unfair contract terms in its Prime subscription contracts, then relying on those terms to introduce advertising to its Prime Video streaming service. The regulator alleges that subscribers were left with no choice but to pay extra to avoid ads, after they’d already paid for an ad-free service, and that Amazon was therefore acting illegally. Amazon is considering is position. Behold the scuba-diving cyborg cockroach Researchers from Singapore’s Nanyang Technological University and Japan’s Waseda University have developed a diving suit for cyborg cockroaches. As described in a Nature paper titled “Underwater Suit-Wearing Cyborg Insect Capable of Hours-Long Diving and Terra-Aqua Travel,” the researchers implant electronic controllers in living insects and use them to induce movements. Not content with re-wiring cockroach brains and steering them around, the boffins built a suit that “integrates an oxygen generator and oxygen delivery tubes” so that cyborg roaches can survive underwater. The scientists think roaches wearing the suits could be sent into small submerged spaces, such as drains filled with rubble, to help with search and rescue missions – or perform some maintenance. The Register looks forward to robo-roaches finding a role in datacenters that employ immersion cooling. ®
9:12
The Coolest Hat At The Hacker Camp
People in hotter parts of the world may permit themselves a grin at this, but Europeans have recently been suffering under an unseasonal June heatwave. Most of us have been cowering inside with our air conditioners, but not [Making Stuff With Mike]. He’s adapting a safety helmet with a Noctua fan for only slightly uncool on-the-go cooling.
On the face of it, the hat is a straightforward hack. [Mike] mounted a 3D-printed chimney to the top of a hard hat and placed a fan in the top of it. But as always, there’s a little more to it than meets the eye, and in this case it’s because he’s modeled the hat/chimney interface by 3D scanning the hat and using the scan to create his CAD model. The two are attached with four small bolts, and a set of large holes are made in the hat for airflow. Taking it out for a spin, he finds it does the job, but has a few ideas for improvements.
So Mike’s ready for the upcoming BornHack hacker camp, which Hackaday has been to a few times. We’re not so lucky with headgear, but at least if there’s a heatwave, they have plenty of hammocks in the trees.
10:07
Kernel prepatch 7.2-rc2
Linus dropped the 7.2‑rc2 prepatch on a quiet Sunday afternoon and the build looks pretty ordinary, more like the recent releases than a tiny incremental bump. The size is a bit smaller than the 7.1 rc2, which suggests the changes are tidy rather than a flood of new code.
What’s under the hood is a collection of routine clean‑ups, a few driver updates and some modest refactoring that keep the kernel humming along. Nothing dramatic, just the kind of steady progress that keeps the ecosystem stable.
The community’s early testing feedback is positive so far, with no glaring regressions popping up. It feels like a solid step forward, giving developers a reliable base to keep iterating on.
If you’re tracking the kernel’s evolution, this rc2 gives you a calm checkpoint before the next wave of tweaks rolls in. It’s a good moment to spin up a test box and see how the latest bits behave in your own setup.
10:53
Hundreds Support Legal Defense for Engineer Charged with Destroying Flock Surveillance Cameras
"Hundreds of freedom lovers are rallying behind a US Air Force engineer" who's been accused of damaging over a dozen AI-integrated surveillance cameras last year and even knocking down their poles.
Long-time Slashdot reader schwit1 shares
this article from Futurism:
According to local channel WAVY, Virginia-based Air Force engineer and mechanic Jeffrey Sovern is facing 13 counts of destruction of property, as well as six counts of both petit larceny and possession of burglary tools related to the destruction of Flock license plate cameras... [Wavy reports the cameras were sometimes pointed in the wrong direction or thrown to the street.]
Armed with garbage bags, spray paint, and even chainsaws, a not insignificant number of privacy vigilantes have taken the fight to Flock, using any means to free their neighborhoods of the ominous surveillance poles. On a GoFundMe page to raise money for his legal defense, the 41-year-old Sovern explained that this kind of privacy-minded vandalism has far more support than would outwardly appear...
Sovern kicked off the campaign late in December of 2025, where he encouraged his supporters to "reach out to the local governments and demand that these systems are taken down." The Virginia resident initially set his funding goal to $8,500. As news of his case has spread across the web, the amount of support has far outpaced those already-hopeful aspirations. [Two hours ago the legal fund stood at $23,326 from over 680 donors].Read more of this story at Slashdot.
12:11
Claude Changed: The July 2026 Way to Use it
Here is the uncomfortable truth about Claude in 2026.
Choosing the best model is now the least interesting part.
You can open Claude, select its most expensive model, write a clever prompt, and still get something you could have produced with a free chatbot two years ago.
Or you can use a cheaper model inside a properly built system and finish work that normally takes an afternoon.
The difference is not a secret prompt.
It is the setup around the prompt.
Claude Chat helps you think. Projects hold the background. Skills remember the process. Connectors bring in your real information. Cowork works across your files and apps. Claude Code builds and tests things. Claude Design creates visual work. Dynamic Workflows coordinate many agents when one agent is not enough.
That is Claude now.
Not one chatbot.
A collection of working environments built around the same model family.
Most people are still using only the first box.
They ask Claude to write an email, summarize a PDF, explain a topic, or clean up a paragraph. These are useful tasks, but they leave almost everything else untouched.
The real move is to stop asking:
What should I prompt Claude?
And start asking:
Where should this work live, what context does Claude need, which tools should it use, what is it allowed to change, and how will it prove the job is finished?
This guide will set up that entire system.
Not as a tour of buttons.
As something you can actually use.
Claude has several layers, and each one is good at a different kind of work.
Claude Chat is where you ask questions, write, analyze, learn, research, and think through decisions.
Projects are separate workspaces containing their own instructions, reference files, and conversations.
Skills are reusable procedures. They teach Claude how to perform the same job consistently without making you rewrite the instructions every time.
Artifacts are finished or reusable outputs such as documents, tools, dashboards, diagrams, calculators, and small applications.
Connectors let Claude work with information inside services such as Gmail, Google Drive, Slack, Notion, Calendar, GitHub, and other connected tools.
Cowork is where Claude operates across files, folders, connectors, browser tabs, and desktop applications to complete multi-step business work.
Claude Code is the technical environment where Claude reads projects, edits files, runs commands, tests its work, and builds software.
Dynamic Workflows let Claude Code coordinate larger jobs across multiple agents and phases.
Claude Design is the visual workspace for creating prototypes, slides, one-pagers, interfaces, and design systems.
The API and Agent SDK are for putting Claude inside your own products and automations.
Anthropic now presents Claude across chat, desktop, browser, terminal, IDE, and API surfaces. Claude Code itself is available in the terminal, desktop app, browser, and supported development environments.
You do not need to use all of them.
You need to know which one matches the job.
Inside the full guide, you’ll get the complete July 2026 Claude setup: which model and effort level to use, the exact Project and prompt system, reusable Skills, Connectors and MCP, safe Cowork workflows, Claude Code commands, Dynamic Workflows, Design and Artifacts, plus practical API routing, cost controls, and approval rules so Claude stops being a chatbox and becomes a reliable working system.
14:56
How to Reduce LLM Costs by 50-60% Using Model Routing
Brian Armstrong from Coinbase said humans shouldn’t be the ones choosing which model answers a request. Model selection is a job for AI.
Coinbase is already running on that belief. They built a routing layer in front of every request.
LLM routing sits between the app and model providers, reads each prompt, classifies the task, and forwards it to the right model automatically.
Coinbase used it to match each prompt to the right model and default to cheaper options where the task allowed it. Cache hit rate went from 5% to 60% and spend ended up nearly cut in half while usage kept growing.
Plano (here’s the open-source GitHub repo) is an open-source implementation of exactly that routing layer, the one that decides which model answers which request, without your application code ever specifying one.
The way it works is that you define routing preferences in plain English, and Plano orchestrator infers the domain and action of each incoming prompt, matches it to the right preference, and picks the model automatically.
Next, let’s set up Plano in front of Hermes Agent, and watch it route requests to different models based on what each prompt actually needs.
Before the setup, let’s understand one key concept.
When an agent runs a task, it doesn’t send a single request. One user prompt triggers several LLM calls in sequence through tool use, each building on what came before.
With a hardcoded model, that’s never a problem. But once routing is in place, each of those calls gets classified independently.
Consider an agent debugging a function. It makes a planning call, runs a tool, reads the output, then makes another call to analyze what came back.
Call 1 routes to claude-sonnet-4.5 for code generation. Call 4 gets classified as analysis and routes to gpt-4o-mini instead.
Two problems occur when that happens.
The model on call 4 has no memory of decisions made in call 1.
And every model switch invalidates the cache, so you’re paying full token rate again from scratch.
Plano solves this with model affinity.
The first call in a session routes normally, and Plano pins that model to a session ID passed in a header.
Every subsequent call with the same ID goes to the same model regardless of how it gets classified.
import uuid
from openai import OpenAI
client = OpenAI(base_url="http://localhost:12000/v1", api_key="EMPTY")
affinity_id = str(uuid.uuid4())
response = client.chat.completions.create(
messages=messages,
tools=tools,
extra_headers={"X-Model-Affinity": affinity_id}
)
The pin lasts 10 minutes by default.
When the task changes, you generate a new ID, and routing starts fresh. For multi-replica deployments, backing it with Redis keeps the pin shared across instances:
routing:
session_ttl_seconds: 600
session_cache:
type: redis
url: redis://localhost:6379
Model affinity handles consistency within a task. Now let’s build the routing layer using Plano.
Plano runs as a local proxy on your local machine. The routing config, model providers, and listener settings…everything is defined in one YAML file.
Plano supports three routing methods:
Model-based: the client specifies an exact model name on every call.
Alias-based: you define a friendly name like
fast-model
orreasoning-model
that maps to a real model behind it, so your app never hardcodes a provider, and swapping models is one line in the config.Preference-aligned: the client specifies no model at all. Plano-Orchestrator infers the domain and action of each prompt and picks the right model automatically based on plain-English descriptions you define.
For alias-based, the config looks like this:
model_aliases:
fast-model:
target: gpt-4o-mini
reasoning-model:
target: claude-sonnet-4.5
We’ll use preference-aligned routing in this walkthrough.
First, we install Plano. Also, we’re using OpenRouter as our model provider, so one key gives us access to every model we need without managing separate API keys:
pip install planoai
export OPENROUTER_API_KEY=sk-or-...
Note: OpenRouter is one option. Plano works with any OpenAI-compatible endpoint, so you can point it at a self-hosted model running on Ollama or vLLM just by swapping the base_url.
Then we create the config.
The routing preferences tell Plano which tasks go where.
