Reading about stranded compute, data centers that are half-built but not operational for any number of reasons. Then I'm reading about the debt burden with neoclouds and how some chunk of peeps think that will implode. And now this micro-neocloud, for lack of a better term, it's literally a VC office offering compute and it's not free. We live in interesting times...

Saw this shared in one of my Slack groups. I like this b/c it's a playful take on 2x2s which I absolutely love when I'm thinking about strategy + positioning with products.

This is an incredible read in that it really highlights what is involved to build S-tier hardware: 9to5mac.com/2026/10/26/apple…

They used to say in software that you will rewrite the codebase every 5 years. What is it today?

TIL, there is something called adaptive high beam. You can keep high beam on but turn off pockets of the beam which dynamically adjust to avoid blinding oncoming drivers. I guess this tech has been available in EU cars for some time but only more recently was approved in the US.

I might sign-up for Uber because of this. I need to walk a lot more!

One consequence of weekly token limits is that some chunk of peeps I know who have grown up w a scarcity mindset feel they have to use 100% each week and so they run 100 agents reviewing every GitHub repo on the planet for no other reason then to burn tokens.

What is the current suggested best practice for dealing with AI bot replies. Is it block? This is for both X and Threads, I seem to get a lot more with the latter.

Not surprisingly, many web publishers are increasingly blocking AI crawlers or agents, especially when they have ad dollars to lose or unique content to protect. I'm eager to see if WebMCP gains adoption. If you're not familiar, WebMCP is a new middle ground where a website can define what an AI agent is allowed to do and expose those capabilities as tools. What's really awesome about this is that WebMCP could become a new API layer across the internet and in some ways, this reminds me of when AJAX was released which ushered the Web 2.0 era.

Am reminded of this infographic we published w my 2 startups ago based on millions of connected calendars

Obvious now that I’m hearing it out loud but usage based pricing isn’t novel, telcos and utilities have been doing this for years.

At a dinner and just heard the acronym, BYOM (bring your own model)

What do hosts think of this AirBnB policy change? TBH, my use of AirBnB has declined but it’s still my default when visiting with a larger group of friends.

Finally got around to reading about Scion, Anthropic’s replacement to Git. Amongst takeaways, the one I’m most happy about is it took 7 engineers, 5 months with Claude. While peeps are trying to vibe code or one-shot apps, real production software still requires real engineering minds.

Mentioned on the SPCX earnings call, "computer per watt" -> that's a fascinating metric, in part, b/c we are normalizing down to raw materials or in other words, if AI commoditizes the app layer, then it's really about the stack underneath (hardware, silicon, energy, raw materials etc).

Glad to know that Facebook isn’t impacted by the PHP vulnerabilities. Thought there might still be some code from ‘04 :)

Seeing a lot of takes on whether Muse is going to succeed. Few comments: 1. Anything with the full distribution force of Meta will perform from a stats perspective, but that’s only half the story (eg Threads). 2. When Siri launched (pre-Apple acquisition), the a-ha moment was incredible. It drove massive virality (and also crazy churn as the promise wasn’t met). Subsequent assistants in that era never achieved the same growth. Without a new a-ha, assistant apps had to operate like standard consumer apps and optimize for things like WOMO (word of mouth optimization) which is super hard since no two consumers can describe their assistant use the same way. ChatGPT went viral because it created a new a-ha moment, but no subsequent assistant has achieved similar macro consumer growth. 3. When working on consumer, I like to map to a core vice. The best vices are things like “make me famous,” “make me rich,” or “help me meet a partner.” For ChatGPT, it’s “make me sound smart.” Muse focuses on getting tasks done, which maps to “save me time.” The challenge is macro consumers aren’t obsessed about saving time (eg productivity). Currently, Muse is tackling harder consumer tasks to drive more “a-ha,” but I am curious to see if that can be sticky enough.

Was reflecting w a college buddy yesterday, how much my Google search usage has cratered. A year ago, there were pockets but today, I often go 1-2 days w/o a single Google search.