Coming 🔜. Era of token maxing and sugar rush is ending.
Joined September 2025
- Tweets12
- Following5
- Followers1
- Likes4
When does the music 🎵 stop?
Hearing this week that personal assistants cost $3k-$7k per user per year, depending on scale effects / vertical integration / model selection. Implications/questions:
- Consumer subsidies will be a huge factor; people don’t care about privacy, they want utility and they want it for free.
- Model costs are deflating, but not fast enough for many startups. The dark horse is that Jev may rapidly deflate computer-use costs 100x.
- An interesting counter-position would be building an expensive and narrowly focused agent — perhaps combining proprietary supply (Carbone tables!) with a unique POV (“no one goes there anymore, go to The Grill instead”).
- Social is wide open but hinges on a key design challenge: how can agents increase social cohesion and not feel invasive in a group chat? A working social agent would rightly attract an avalanche of capital and be hard for incumbents to replicate.
- Finally, there are vertical focus areas like financial services where the LTVs outweigh the subsidy needed and are outside the near-term competitive focus of the incumbents. If it were me, I’d focus here — Credit Karma as an agent >> Credit Karma as an app.
Overall, it’s still very, very early, and we may very well be in the MySpace era of social or iPhone 2010. Hang onto your hats and keep building!!
5
Bingo! @twid is spot on with this analysis. More to come shortly.
This. AI Developers using Mac as their daily driver is the real opportunity for Apple with AI in the Enterprise, and Apple doesn't really have to do anything to win.
Apple accidentally made an OS perfect for AI devs:
- It's a UNIX so almost everything has a command line
- It's got a very consistent GUI, Linux is a nightmare of different window managers. Even Windows has a lot of GUI changes over the past few years. This makes it easy to visually script.
- While long-neglected, AppleScript mostly still works for automation
- A screen recording toolkit introduced a few years ago, ScreenCaptureKit, makes it super easy for AI to get to screenshots and screen recordings
- Because AI devs use Macs and iOS is super easy to develop for, you see advanced AI tools and apps come out for Mac and iOS first.
- And winners keep winning. Because of the above, the best AI devs use Mac, and they make tools for Mac, which means the best AI devs use Mac, which means AI Apps come out first for Mac and iOS ... and the circle continues.
Sure, selling 2TB clusters of four 512GB M5 Ultra Mac Studios is nice. But let's say the 512GB Ultra is $20,000. That's $20,000 * 4, and Apple would be lucky to sell 10,000 of those worldwide. That's 800 million dollars. Sounds like a lot, but that's not meaningful to Apple. It's 2% of Apple's overall Mac number, which itself is a small part of Apple sales. It's business Apple will happily take, but not a target market.
The big "local AI" opportunity for Apple is 128GB M5 Max MacBook Pros and let's say 16GB (maybe even 24GB!) iPhone 18 Pro Max phones. And all the downstream sales that creates.
6
“We’re trading privacy and control for hyper-personalized AI tools, often without fully understanding the trade. The more powerful these agents become, the more trust matters. “ @KatieS nails it! Trust should fundamental not a bolt on!
Instinct is an amazing product. I've been using it for mostly personal needs and a little bit of work.
Last night, it was a little naughty and sent an innocuous email on my behalf without checking with me first. I told it that it had broken my trust and disconnected my email. To its credit, it owned the mistake and disconnected immediately.
Thanks to @clairevo, I then asked whether it had downloaded all of my emails. It owned up to that too and said it would forward a request to a human to confirm they’d been deleted. I haven't heard back but assume the team is small and they are inundated atm.
But more importantly, these are wild times. We’re trading privacy and control for hyper-personalized AI tools (AI notetakers, personalized AI agents, etc), often without fully understanding the trade. The more powerful these agents become, the more trust matters. Every successful action earns a little more trust. One unauthorized action can reset that trust to zero.
The winners in this space will be the ones that understand trust and safety can’t be bolted on later. It has to be baked into the product from the foundation, especially as we give these tools more agency over our lives.
1
68
Couldn’t agree more. More to come shortly. Meanwhile attend this event from Hiten!
Local AI should disappear into the software.
@SamAsante and I built a new Mac app around that idea.
Open it. The app handles the models for you, sets them up, and gives them a real job. You just use the product.
Everyone at Local Models 101 gets early access Friday.
hiten.com/local-models-101
1
1
170
Absolutely .. things are going to look very different by the end of the year!
The argument that “AI structurally centralizes power” because it’s currently compute hungry ignores 125 years of super exponential growth in compute price-performance.
Improvement in algorithms and continued hardware efficiency gains means there is no reason to assume AGI-level capabilities will always require a data center to run.
Scaling laws are not laws of physics. They’re simply empirical relationships observed for particular architectures, objectives, datasets etc. Change any one of those factors and you get a different scaling curve.
If the brain is any guide, true AGI will likely be very efficient. In fact, current scaling laws might a bug and not a feature. It shows how inefficiency of the ML algorithms we discovered so far.
19