Hanami

I like cats. I like jogging.

[Rohan Paul @rohanpaul_ai]

CNBC: Moonshot AI, the Chinese lab behind Kimi, has sought to rent Nvidia Blackwell chips in European data centers.

It discussed deals with neoclouds, non-Chinese GPU rental firms that can buy the hardware.

U.S. export controls bar shipping those chips to China but don’t broadly cover renting them from servers abroad.

The White House accused Moonshot in July of using the same route in Thailand, and a source said the European talks came afterward.

One person described clusters of 8,000 or more Blackwell GPUs, sought mainly for serving Kimi to users.

Under one proposed lease, Moonshot would pay 30% to 50% more upfront than a non-Chinese renter, cushioning neoclouds if Washington bans remote access.

Nvidia reportedly told cloud providers that on new contracts, at least 51% of services must go to non-Chinese customers.

That cap still leaves 49% for Chinese buyers, letting a large European neocloud rent Moonshot 8,000 GPUs.

[Nicolas Bustamante @nicbstme]

The new UX is the API. The new user is an agent. For decades, the best product builders were the ones who understood humans better than anyone else.

Building great software meant having a really good intuition for UX. Will people know where to click? Can they find the right action? Is the product intuitive?

A lot of the best builders developed that intuition as teenagers building websites, then spent years learning what makes software easy to use.

But now we're building for agents, and it's a completely different skill set. Take booking a business trip. For humans, you build a nice interface to search flights, pick a hotel, add things to your calendar and submit expenses.

For an agent, none of that matters. What matters is whether it can figure out which APIs to call, understand the parameters, handle errors, and complete the whole thing without getting stuck halfway through.

An API can make perfect sense to an engineer and still be terrible for an agent. So for example, imagine an API where creating a calendar event, inviting attendees, booking a meeting room and setting permissions are four separate operations. Perfectly logical from an engineering perspective. But now an agent has to discover the right sequence, understand dependencies, handle partial failures and clean up if something goes wrong. Why not let it express the entire intent in one call and have the system handle the rest?

I think we're going to see a new generation of product builders who have incredible intuition for how agents think, what confuses them, and how to make software that they can actually use.

It's a new kind of product empathy (let's call it agent empathy, haha!)

[Hanami @CatHanami97880]

@mori_notes Works like magic!

[Mori @mori_notes]

Me testing our phone agent with long nails :)

Meet 𝐌𝐚𝐭𝐞𝐨. My team and I can finally show you what we've been working on. 🫡

𝐈𝐭 𝐮𝐬𝐞𝐬 𝐚𝐩𝐩𝐬 𝐨𝐧 𝐚 𝐬𝐞𝐩𝐚𝐫𝐚𝐭𝐞 𝐯𝐢𝐫𝐭𝐮𝐚𝐥 𝐬𝐜𝐫𝐞𝐞𝐧 𝐨𝐧 𝐲𝐨𝐮𝐫 𝐀𝐧𝐝𝐫𝐨𝐢𝐝 𝐩𝐡𝐨𝐧𝐞, 𝐬𝐨 𝐲𝐨𝐮 𝐜𝐚𝐧 𝐤𝐞𝐞𝐩 𝐮𝐬𝐢𝐧𝐠 𝐲𝐨𝐮𝐫 𝐩𝐡𝐨𝐧𝐞 𝐰𝐡𝐢𝐥𝐞 𝐢𝐭 𝐡𝐚𝐧𝐝𝐥𝐞𝐬 𝐚 𝐭𝐚𝐬𝐤. 𝐘𝐨𝐮 𝐜𝐚𝐧 𝐜𝐡𝐞𝐜𝐤 𝐨𝐧 𝐢𝐭 𝐢𝐧 𝐚 𝐬𝐦𝐚𝐥𝐥 𝐩𝐫𝐞𝐯𝐢𝐞𝐰 𝐨𝐫 𝐬𝐭𝐨𝐩 𝐢𝐭 𝐰𝐡𝐞𝐧𝐞𝐯𝐞𝐫 𝐲𝐨𝐮 𝐰𝐚𝐧𝐭.

Getting help shouldn't mean waiting around for my phone back.

It's our very first version, and I'm glad we can finally show it. Stay tuned for more updates! 🥳

[David Pawlan @DavidPawlan]

Siri seems to be shifting into an agent platform, feels like a deviation from Apple’s historical closed ecosystem

A TLDR of what’s available:

> Any app can now give Siri its actions, like ordering, booking or tracking, and Siri runs them inside the app

> Devs can configure Siri to chain several actions from one ask

> It understands what's on your screen and your personal context so you can ask Siri to “send ‘this’ to David” without defining what ‘this’ is

> It can find anything an app indexes, not just Apple's preset types

Developer beta access is released today with public access coming in a few months

This will be game changing for the agent revolution

[David Pawlan @DavidPawlan]

the simple fact is that when people use AI to shop, it's because they're craving a single answer as they arrive with the intent to purchase, not to peruse

this is why i'm so bullish on agentic commerce

it's the antithesis of window shopping

[Jordan Jenkins @jordan1oi]

gpt防御性笑话发言大合集🤣

[WeChat @Weixin_WeChat]

Your AI agent just got teammates. And a schedule.

WeKnora v0.9.0 brings multi-agent collaboration and scheduled tasks to our open-source AI knowledge platform.

🤝 Share the workload: break complex tasks into independent subtasks, delegate them to agents and bring the results together.

⏰ Put tasks on a schedule: set them to run once or repeat automatically, with execution history and results to review.

Explore the release on GitHub: https://t.co/8oFPodiR8Q

[Noah Shinn @noahrshinn]

Check out anywhere with Shop Pay

Today, we’re expanding our partnership with @Shopify beyond finding, buying and tracking orders from Shopify stores. Now, your Instinct can use Shop Pay at any online store.

The card and address you've already saved in Shop Pay now work across stores that aren’t on Shopify, too. This makes it easy for your Instinct to make secure purchases on your behalf.

You approve every purchase in Shop Pay, and your card details never leave Shop Pay.

[Adam KP @adamkp]

okay, this is actually the first legit use case for a touchscreen MacBook

[Hanami @CatHanami97880]

A small Chrome extension: labels every tweet by topic and folds those you don't read: ads, spam, repeats, politics, crypto, for free. https://t.co/x8YC1kSBq5

[Hanami @CatHanami97880]

@SidChatterjee @Cloudflare Finally!

[Hanami @CatHanami97880]

@DaveRBanerjee So there is a bigger potential market here for insurance. Byebye AI bubbles.

[Hanami @CatHanami97880]

@dbreunig Basically you can prove any assumption about LLMs if you try hard enough.

[Hanami @CatHanami97880]

@firesidealpha I need affordable models, but they just talked their profits.

[kadd @kaddisdeployed]

Nanbeige just dropped a 33B open agentic coding model worth watching

it is Apache 2.0, dense, and carries a 512K context. On their card it posts strong numbers against other ~30B models on terminal and software-agent tasks (Terminal-Bench 2.1 at 80.9, DeepSWE at 65.5). It trails slightly on SWE-Bench Pro

Practical notes from the card:

- BF16 weights are ~66 GB
- Recommended inference is multi-GPU (tensor parallel 4) with vLLM or SGLang
- Tool calling is supported (XML preferred)
- Thinking mode can be toggled; they recommend keeping reasoning for agent work

another open coding agent to put on the watch list

especially if you want something you can run and fine-tune yourself

[Tak 🦞 @cherry_mx_reds]

Anthropic is now recommending that companies REMOVE unnecessary subagents from their AI systems.

This is a pretty significant shift in agent architecture, with their team finding that a single agent equipped with skills can frequently outperform complex multi-agent systems, and one customer seeing roughly 2x faster execution.

The emerging rules of thumb:

→ Skills: Teach the agent HOW to do something, including procedures, expertise, and instructions.

→ Tools / MCP: Give the agent the ability to DO things through APIs, actions, and live data.

→ Subagents: Use for parallel work, isolated context, or tasks a single agent struggles with.

The interesting part is progressive disclosure. Instead of spawning specialized agents or stuffing everything into a giant prompt, one agent discovers and loads skills and tools as needed.

Their advice is to fold existing subagents into the main loop, turn their instructions into skills, expose their actions as tools, and run evals to see if performance holds up.

As models have gotten smarter, improving agent architecture is becoming less about what you can add and more about what you can delete.

https://t.co/DNq60Skkmc

[Mustafa Ali @mustafa01ali]

Local machines had become a bottleneck for agentic mobile development at @Shopify, so we gave agents their own remote dev environments. They now author 61% of our merged PRs 🧵