One person with the right tools is now enough
The world is changing. We have incredible new tools at our disposal. Humans who combine their creative ingenuity and thinking with the skill set of wielding these tools effectively now have the ability to achieve great things. Each of you is a Sparta warrior but now you don't need 300, 1 of you is enough because you have an unfair advantage of having the weapons of Gods in your hand. Go make history!
- Plank
Keep reading
Newer: The era of small models in the enterpriseEnterprise AI
AI's impact beyond coding will take time to diffuse
A majority of the impact of AI so far has been in the software engineering and coding domain, and this is yet to make an impact on the broader economy. As an example, so far, Waymo was more expensive than Uber and Lyft, but recently, in my experience in SF, they are now consistently ~25% lower. The same will happen in other domains, but will take time for the diffusion because the bar for an AI system to replace a human process is much higher. We expect a much higher accuracy from automation than from doing the work ourselves. That requires solving a lot of edge cases, which is tedious and takes time. The problem there is again not how to solve it, but to understand what are those edge cases and the context of how these systems interface with the real world.
Models and researchAI agents
Local models versus labs is not a binary choice
I think the debate of local models vs labs is not a binary debate. Any reasonably complex platform will use both. The key is how do you decide what agents you use local models for and what agents you use GPT/Claude for.
Plank
Beyond Perfection
Beyond Perfection Out beyond the boundaries of excellence and perfection, There is a place. I’ll meet you there. Where the highest standard is merely the beginning, And greatness no longer seeks to be measured. When the soul arrives beyond what was thought possible, Even the word perfect loses its meaning.
Software architectureCoding agents
The joy of meta programming with AI tools
One of the most joyful things for me is meta programming. And thanks to the new AI tools, it makes it super easy to just think in the meta space and not have to worry about the implementation. And as long as you have your meta functions layered correctly in the concept space, the actual implementation by AI doesn't even matter since it's forced to be correct at the input/output invariance layer. FUN!
Software architecture
Happy Friday: what are you shipping today?
Happy Friday! What are you shipping today? I remember that many years ago, the conventional wisdom was that you never ship to production on Friday. The times have changed. Now you ship to production every day and every hour. The world has never moved faster. LFG
Plank
Sequoia, relentless execution and a golden era to build
Interesting to see Sequoia Capital continuing to innovate in Venture Capital. They have always set a high bar in how you operate with high integrity and rigor. They set a very high bar for themselves and come with a mindset of relentless execution. We are living in a golden era of innovation, it is the time to build and if you are someone who is excited to build your version of the future, now is the time. The mindset that is needed to win is that every week, every day, every minute is critical. What did you get done in the last 5 minutes?
Enterprise AI
Safe, sovereign AI deployment is the decade's top challenge
Safe secure sovereign deployment of AI will be the highest value challenge of the coming decade. There is an opportunity to rethink the security, deployment and fundamental business paradigms in this domain. The key is to get unshackled from the way of thinking of the past era. We have to start by accepting some new tenants like the cost of software is zero, cost of tokens in enterprise is near zero, model capabilities are already at AGI level when it comes to enterprise, and security and sovereignty of data is unsolved at the moment. It is an exciting space to innovate in.
Models and research
GPT-6 Astra edits Qwen's weights to make it better at math
Wow! This is like science fiction. This might be the clearest sign that we are on the verge of RSI (Recursive Self Improvement). In this case, GPT-6 Astra is able to modify the weights of an open-source Qwen 3.5 model to make it better at math by just editing 0.016% of the weights. Link to the paper and analysis in comments Paper: rikiyat.github.io/blog/weight-space-programming Analysis and reproduction of the result from Simon: simonwillison.net/2026/Nov/3/qwen-can-interpret-mod…
