More AI More Bubbles More Money
This month was all about where the year of agent ends. Family, what it means to have a large one. Existential crisis.
Derivative Hot Takes (Again)
The Year of Agents was a hit and miss in 2025 in terms of adoption rate. But the story will be very different in 2026.
Hardware optimizations are plateauing and the moonshot projects are the only real hope for Google and others. I am speaking space tech.
Microsoft finally cleaned its house and got it together. The Copilot ecosystem now makes robust sense, Foundry, Work and other agents now interact meaningfully. They call context engineering their IQ stack.
Interesting corners of the internet
Jeff Bezos is BACK baby. He is now the co-CEO for a new AI startup - Project Promotheus. https://www.nytimes.com/2025/11/17/technology/bezos-project-prometheus.html
Da Bombs from China just ain’t stopping. In an NYT article that details how various AI startups have created various surveillance softwares from tracking people, their habits, interactions, online-shopping, medical history etc predicting a person’s state of mind to speech recognition models that target minority group dialects. Making Chinese startups the hub of AI +. (More on it at https://nationalsecurity.virginia.edu/research/chinese-communist-partys-layered-artificial-intelligence-strategy
Netflix bought Warner Bros ?!
Australia bans social media for under-16s and they (in typical teenager fashion) find workarounds.
Models released since I last wrote a newsletter
Devstral 2 & Mistral 3
Deepseek 3.2
ChatGPT 5.1 & 5.2
Kimi K2
GLM 4.6
Qwen 2.5 & Qwen Max
Llama 4! (Ps this one’s probably a bummer. Mark decided to go the closed route after Yan left Meta. YOU READ RIGHT.)
Phi-4-Reasoning
InternLM-3
Gosh, gotta catch my breath.
Philosophical bullshit
We, humans, are perhaps the only species, capable of separating events from existence. We coined this ‘experience’. The ‘human’ experience. And in pursuit of understanding this separation, we keep diving deeper. For example, on reading a sentence “Stars wrinkled like old paper as the cruiser punched a hole into the unknown.”, my first thought was, how long does novelty last in the human brain? in the human interpretation of our surroundings? of ‘human experience’.
Interesting reads
AI Engineering is a page-turner so far. The diagram of a Transformer model architecture below is the cleanest and simplest one I have seen.
No nerd corner this week to deep dive into the details of that model.
I read the paper Hidden Technical Debts in Machine Learning Systems published by Google in December 2015.
My key takeaways are:
Always consider the cost of convenience (so true in today’s age of LLMs.)
Developing and deploying ML systems is relatively fast and cheap but maintaining them and building them over time is the difficult part.
There’s a very interesting diagram that tells you just how much of the actual work is software engineering vs ML code. (Back in 2015 - Newer libraries might have changed the ratio but the work still remains the same).
Entanglement - This is where they emphasize everything is linked. CACE - Change Anything Change Everything. This is different from software systems, for example, changing input parameters, hyper parameters or features changes the way the entire system behaves.
There’s also a whole part dedicated to downstream data dependencies. Especially the undeclared ones.
Underutilized data dependencies or not-so-relevant data dependencies accumulated over time. These can be managed by static analysis?
What the paper does not focus a lot on is the different data capturing mechanisms that need a large maintenance effort and fine-tuning over time.
Books corner
I read the book How To Make a Life, it took me two years! Because of the amount of trauma embedded in it.
It is about intergenerational impact.
Each character’s pain is inherited, reshaped, or misunderstood by the next generation. I wonder what trauma I carry from my own experiences?
The book isn’t just about a family. It’s about what families give and cost.
Takeaway - family isn’t just about warmth and belonging, it’s about inheritance - not of money, resilience and responsibilities.
It’s about making your life worse at times to make the family better. And battling the consequences and events life throws your way as a group. This is a gripping read for people like me, people that wonder what it would be like to have a large family.
Personal wins
This is a new section where I will list everyday wins that I am proud of. The catch is, the win should be so ordinary that anyone can do it.
Worked through a illness but balanced it with equal rest this week. Did not feel guilty for skipping workouts in this timeframe.
Was extremely vulnerable with a friend.
Observed a friend being selfish in petty things and forgave them for it without thinking twice.
To read next
https://claude.com/blog/making-claude-a-better-electrical-engineer
This piece really made me think, what if the amazing new models like Llama 4 found open, ethical paths insted of closed routes, democratizing AI for good?