The future shape of software
The thesis is that as models get better, protocols standardize, and agent runtimes become more capable, the center of gravity in software shifts away from standalone UI-heavy apps and toward services, automation, execution quality, and deep domain infrastructure.
Some of those services will still present themselves as branded apps. Some will be invoked through voice. Some will be reached through a marketplace. Some will be called through CLI-like execution surfaces or protocol connectors. Some will be embedded invisibly inside another company’s product.
The visible UI layer becomes thinner and more replaceable; the execution layer, the trust layer, and the domain-quality layer become more important.
The richest guy on Earth SHOULD be the guy making cutting edge cars and rockets instead of dudes who sell purses and perfumes or dudes who run investment firms or dudes who made Facebook.
Kimi 2.6. Chinese labs seem to have built image models that are more focussed on the task and truth rather than trying to make the user happy. Better way to design.
I expected Grok to also say no to me - but it failed. Grok generated the least creepy photo though.
The internet is a strange place. Our models have picked up some unusual things and so they fill in gaps with even more peculiar information.
One model - denied to generate an image. 👇
a very important thing, also shown here, is how OpenAI's costs are actually lower with GPT5.5 xhigh vs Claude which is 'cheaper'
Anthropic has the worst harness and design and Claude consumes & produces garbage tokens that serve no value except helping Anthropic increase their revenue.
@AodenTeoMT Hey, I have to be really honest here. The product sucks. The model hallucinates and invents words and tones the moment you add any random punctuations or do anything interesting with it.
The demos are cool, but the product clearly not.
If a human employee repeatedly lied to my face that a task was done when it was completely broken, they’d be fired on the spot.
But with AI? I’m stuck paying the damn thing more tokens to confidently gaslight me, and then getting billed again just to have it fix its own garbage code.
I see why Anthropic makes so much money.
Came to Dubai to pick my DGX spark.
Expanding my local compute, running multi-agent harnesses and building something really cool.
Will share more in a detailed blog early next month.