franzi

slop cannon orchestrator

[shift @joinshiftX]

Today, we're launching shift. We're starting by cleaning your apartment in New York City, for free.

Here's how it works. Book a shift cleaning. A vetted shift operator comes to your home wearing one of our devices. They clean. They leave. You pay nothing.

In exchange, we record the cleaning. Robotics is being built on data about how people do daily tasks, and the value of that recording is what funds the service. Anything personal in it is anonymized before the recording is processed.

By now, you have heard about the shift to AI more times than you can count. About the shift toward you, the part where you actually feel it, you have heard almost nothing. Shift is what starts to make it concrete, in specific cities, with specific services.

Today, cleaning in New York. Soon, handymen, repairs, and errands across the globe. And this is just one side of shift, with more on the way.

Comment “shift” and we’ll send you an early access link.

[GitHub @github]

We are investigating unauthorized access to GitHub’s internal repositories. While we currently have no evidence of impact to customer information stored outside of GitHub’s internal repositories (such as our customers’ enterprises, organizations, and repositories), we are closely monitoring our infrastructure for follow-on activity.

[Melissa Chen @MsMelChen]

Singapore’s Foreign Minister, Dr Balakrishnan casually explaining how he built his own AI agent (a 2nd brain for diplomacy) using Claude & WhatsApp integration etc. on a Raspberry Pi

“You cannot govern a technology you have only been briefed on.” 🇸🇬

[franzi @shdli]

Great place to hang cctv camera 🤔

[franzi @shdli]

Why did they add cameras to AirPods? Who wanted that

[ClaudeDevs @ClaudeDevs]

How do you keep Claude working until the job is done? Claude Code helps with this in a few ways, including one we shipped recently: /goal.

[Luc Perkins @lucperkins]

Under-utilized Nix feature: on-the-fly shell environments

[Maxi @AllForProgress_]

On a desk somewhere in Singapore, a small computer not much bigger than a deck of cards has been running continuously for the past several weeks.

It is connected to its owner's WhatsApp, Gmail, calendar, and a personal library of saved speeches and articles. It transcribes voice notes locally. It runs vector embeddings (the technique that lets an AI find what is relevant in a great big stack of stuff) locally.

It maintains a knowledge graph of every fact, every counterpart, and every piece of negotiating history its owner has ever recorded into it. It is, in short, a perfect mind-out-of-mind. Its owner calls it "NanoClaw, a second brain for a diplomat."

The owner is Dr Vivian Balakrishnan, and he is the Foreign Minister of Singapore. He is a trained ophthalmologist, a Fellow of the Royal College of Surgeons of Edinburgh, and one of the very few sitting cabinet ministers anywhere in the world who writes his own computer code.

My kind of guy. And, I think, everyone's kind of guy.

On 21 April he published the entire architecture to a public GitHub gist. He did not announce it through his ministry. He did not spin it into a productivity-tool startup. He wrote it up the way a working software engineer writes up a side project, and put it in the open. The hardware cost is approximately £80; the running costs are between £5 and £20 a month.

I would invite the reader, gently, to attempt the exercise of imagining a single member of the present British cabinet doing the equivalent. Imagine Yvette Cooper publishing the architecture of a system that tracks her foreign counterparts' priors and negotiating history, on a personal GitHub account, with API keys segregated through a credential proxy because she takes her department's data sovereignty very, very seriously. Imagine David Lammy shipping a working tool that drafts his speeches against the record of his own past statements - haha, no, no, stop! I know, it's beyond the absurd.

Then consider the gap.

This is what your public servants ought to be able to do for themselves. This is the class of governing official Britain could have, being not at all absent the talent capital needed. But Dr B's excellence shows that, while we don't lack the talent, we have been getting recruitment all wrong.

Britain has been recruiting senior public servants for two political generations on the basis of who can speak well at a Tuesday lunch club, not on the basis of who can deliver a working result when everyone's back to work. The country that built the institutions everybody else copies built them with people who could think, and write, and do. We optimised on the first two and stopped caring about the third. The third turns out to have been the one that mattered most.

But that's the golden triptych - people who can think, write, and do. Imagine how much better our national fortunes will be when it is people of that calibre at the helm?

It is coming. It Can Be Done.

[Peter Girnus 🦅 @gothburz]

I have three monitors on my desk. The left one shows the order book. The middle one shows Truth Social. The right one shows the investigation queue.

On April 21st, the left screen moved first.

I am a Senior Surveillance Analyst at a commodities exchange. I have held this position for nineteen years. My job is to monitor trading activity for suspicious patterns and generate compliance reports. I am employee of the quarter. I have a mug.

At 19:54 GMT on April 21st, someone placed 4,260 sell orders on Brent crude futures. They did this during post-settlement. The window after the market closes when daily volume is typically in the dozens. Sometimes single digits. Sometimes I watch the screen and nothing happens for forty minutes and I think about whether my daughter is happy.

On April 21st, someone placed $430 million in directional bets in 120 seconds during that window. One hundred and twenty seconds. I timed it on my watch because the system clock rounds to the nearest minute and I have found, in nineteen years, that precision matters to no one but me.

At 20:10 GMT, the President posted on Truth Social that he was extending the Iran ceasefire.

Brent dropped from $100.91 to $96.83.

I flagged the trade. I flag a lot of trades. I want to tell you what happens to my flags.

My flags go into a system called TRACE. Trade Review and Compliance Evaluation. I did not name it. The system generates a report. The report goes to a committee. The committee has a name I am not allowed to share but I can tell you it meets quarterly and the conference room has a credenza with bottled water that is sparkling because someone once put still water in the room and a managing director sent an email about it that was longer than most of my surveillance reports.

The committee reviews my flags. The committee has reviewed all of my flags. Here is the complete record of actions taken on my flags in 2026:

Reviewed.

That's it. "Reviewed" is a status. In compliance, a status is the absence of an action that has been given a name so it looks like one.

Let me show you my flags.

March 9th. Someone bet millions on oil falling at 18:29 GMT. Forty-seven minutes later, a CBS reporter posted that the President said the Iran war was "very complete, pretty much." Oil dropped 25%. Forty-seven minutes. I flagged it.

March 23rd. Someone sold 5,100 lots of Brent and WTI crude futures between 10:49 and 10:50 GMT. Fourteen minutes later, the President posted on Truth Social about a "COMPLETE AND TOTAL RESOLUTION" to hostilities. Oil dropped 11%. Over 13,000 contracts traded in sixty seconds after the post. Fourteen minutes. I flagged it.

April 7th. Someone established a $950 million short position in oil futures at 19:45 GMT. Three hours later, the President declared a two-week ceasefire. Nine hundred and fifty million dollars. I flagged it.

April 17th. Someone placed $760 million in bearish bets twenty minutes before Iran's foreign minister confirmed the Strait of Hormuz would reopen. Seven hundred and sixty million. I flagged it.

April 21st. The $430 million. Fifteen minutes. I flagged it.

That is $2.1 billion in directional oil bets in April alone. Every one of them landed on the correct side of a presidential announcement. Every one of them was placed in a window so narrow you could measure it in bathroom breaks. I flagged every single one.

The CFTC chair told a Congressional committee that his organization has "zero tolerance" for fraud and insider trading. I wrote that quote on a Post-it note and stuck it to my right monitor. The one that shows the investigation queue. The investigation queue has not moved since March.

Zero tolerance. Zero staff. Zero budget. Zero prosecutions under the STOCK Act since it was signed in 2012.

Fourteen years. The law has existed for fourteen years and has been enforced zero times. In compliance, we call that a compliance rate of one hundred percent. No cases filed means no cases lost. You cannot fail an audit you never conduct. We call that excellence.

Last month the White House sent an internal email to staff. I was not on the distribution list but I have read reporting on it and I need you to sit with what I am about to say. The email instructed White House staff not to use insider information to place bets on prediction markets.

The White House had to send a memo telling its own employees not to insider-trade.

I want you to read that sentence again. Not because the instruction was unclear. Because the instruction was necessary. Because someone in the building looked at the same pattern I have been flagging for months on my three monitors and decided the appropriate response was an email.

The President's son sits on the advisory board of Kalshi. He is an investor in Polymarket. Both are prediction markets. Both saw accounts created days before U.S. military action.

One account. I cannot stop thinking about this account. It was called "Burdensome-Mix." It was created in December. On January 2nd, it placed $32,500 on Venezuela's president being removed from power. On January 3rd, Maduro was seized by U.S. special forces. Burdensome-Mix collected $436,000. Then it changed its username. Then it disappeared.

One account is a coincidence. But there were six.

Six accounts were created on Polymarket in February. All bet on U.S. strikes on Iran by the 28th. When the President confirmed the strikes, the six accounts collected $1.2 million between them. Five of the six never placed another bet. The sixth went on to correctly predict the ceasefire date and made another $163,000.

My surveillance system logged all of this. My system logs everything. My system does not have opinions and neither do I. I generate reports. The reports go to committees. The committees meet quarterly. Between meetings, the windows get shorter and the bets get larger.

March 9th: 47 minutes. March 23rd: 14 minutes. April 17th: 20 minutes. April 21st: 15 minutes.

The window is compressing. In March, you had time to make coffee between the trade and the announcement. By April, you had time to send a text. By summer, at this rate, the trade and the announcement will be the same event.

The spokesman said any implication that administration officials are engaged in insider trading is "baseless and irresponsible reporting."

Then the White House sent the email again.

I have been in compliance for nineteen years. I have seen insider trading run out of strip mall offices by men who could not spell "derivative." I have seen pump-and-dump schemes coordinated over WhatsApp by people who used their real names. I have seen a man try to manipulate soybean futures from a Panera Bread.

I have never seen $2.1 billion in perfectly timed trades across five presidential announcements in a single month go uninvestigated.

But I have also never seen a compliance system work this beautifully. Every trade flagged. Every report filed. Every committee briefed. Every quarterly meeting attended. Bottled water: sparkling. Minutes: distributed.

Zero prosecutions.

As long as the flags go up and the cases don't, my performance review says I am meeting expectations.

I am meeting expectations. The system is meeting expectations. The $2.1 billion is meeting expectations. The fourteen-year-old law with zero prosecutions is meeting expectations.

The left screen moves. The middle screen moves. The right screen stays perfectly, immaculately still.

In my field, we call this price discovery.

[SpaceX @SpaceX]

SpaceXAI and @cursor_ai are now working closely together to create the world’s best coding and knowledge work AI.

The combination of Cursor’s leading product and distribution to expert software engineers with SpaceX’s million H100 equivalent Colossus training supercomputer will allow us to build the world’s most useful models.

Cursor has also given SpaceX the right to acquire Cursor later this year for $60 billion or pay $10 billion for our work together.

[franzi @shdli]

@BettyWhoYou @Amtrak same message on 4/10/2026 - the site is being shipped broken to production 🤦

[franzi @shdli]

Banksy and Satoshi—the two most elusive and enigmatic figures of the last twenty years—unmasked within weeks of each other 🤔

[Graham Christensen @grhmc]

I'm so proud for Nix and @github: Dependabot version updates now support the Nix ecosystem, via flake updates. Extra proud that they link to Zero to Nix, our love letter to flakes.

[Home Office @ukhomeoffice]

Police time will no longer be wasted investigating legal social media posts, freeing up officers to patrol the streets and tackle real crime.

By scrapping Non‑Crime Hate Incidents, we are balancing the protection of vulnerable communities while respecting free speech.

[franzi @shdli]

Cursor and Lovable ads on the tube

[franzi @shdli]

why does the high street need to look like this?

[franzi @shdli]

@mikehogan_ 402 This request requires more credits, or fewer max_tokens. You requested up to 32000 tokens, but can only afford 30603. To increase, visit https://t.co/TJRmOsBuZc and add more credits

[franzi @shdli]

I’m spending my days flitting between agents in tmux, trying to manage context through rapid switching and task managers to keep constraints and precedent straight. Context switching is costly, but this new async way of working demands it. There should really be an agent WIP limit.

AI agents may scale output, but supervision doesn’t scale linearly. At some point one engineer managing many agents becomes the supermarket self-checkout problem: too many terminals, one supervisor, rising errors and falling throughput.

There’s no way you can increase output indefinitely and still hold the full context with the precision needed for work you’d actually stand behind. Software in critical applications need pass the keystone test: the builder should be willing to stand underneath it.

[franzi @shdli]

Finding it very powerful to delegate tasks straight from things / omnifocus to agents