In early 2020, I wrote a couple of posts here about what AI could do for society. A few weeks after the second one, Apple’s AI Strategy lead John Giannandrea stood in a room in London and said “we haven’t even seen the beginning of AI.”

He was right, and none of us really knew how right at the time.

Since then I completed an MSc in AI, spent several years building production systems at AI startups and as a Generative AI Engineer, and watched the field transform in ways that were thrilling, humbling, and occasionally alarming. I’ve been heads down in the work rather than writing about it, and I’m reviving this blog to change that.

The plan is to build things, document the process honestly, and write about the engineering decisions and tradeoffs as they emerge. The first thing I want to tackle is a question I’ve been sitting with since 2020: what would the architecture of a truly unified AI actually look like, and could we build it today? By that I mean a system that knows your context, connects your tools and services, and acts on your behalf across different parts of your life, something I first sketched out in my 2020 post but that we now arguably have the technology to architect properly.

More on that soon.

Sameen