Notes
Quick thoughts, links, and pictures. RSS
I was recently chatting with a colleague about AI adoption. I mentioned that it’s circular. At first, the leading edge is just a demo, and people claim it can’t do real work. Then it gets better, and people say, well, it can’t scale. Then it gets better, and it scales, and then people say, well, it’s too expensive. Then it gets better, and it’s suddenly not expensive, and it starts again on the next thing. We are seeing this play out with AI a lot at the moment.
Image set1 image · open to inspect
Our nation will come to greatly regret gambling laws changing and the ease and gamification of it on mobile devices. This will destroy many young people, especially.
x.com
Charlie Bilello (@charliebilello) on X
Total amount wagered on sports in the US...
2025: $165 billion
2024: $149 billion
2023: $121 billion
2022: $93.7 billion
2021: $57.5 billion
2020: $21.5 billion
2019: $13.1 billion
2018: $6.6 billion
Boris is making a case for new roles in a collapsed function post-AI workplace.
x.com
Boris Cherny (@bcherny) on X
As engineering, product, design, DS, etc. melt into a new kind of role, I was reflecting on what roles might look like in the future. For example, when I look at the Claude Code team I see what I think is five archetypes:
1. Prototyper: comes up with brand new ideas; churns out
I don’t think there are more than 5 songs from the last 25 years of Christian worship music publishing that are likely to endure. Modern Christian Music is not Lindy.
Probably the most telling experience I’ve seen over the last 25yrs has been the removal of the song books and hymnody of the last 200yrs. Since really the 90s on, it’s been a large push to modernize music. It seems to have failed.
In our church there has been a routine of singing modern worship songs followed by one or two more traditional songs. The audience response is palpable. It’s an immediate difference.
It’s not my goal to disparage attempts at new songs but the systems in place to do it are largely not working.
I suspect in another 30-40 years we will still know the songs of the 1800-1900s far more than those of the early to mid 2000s.
Probably the most telling experience I’ve seen over the last 25yrs has been the removal of the song books and hymnody of the last 200yrs. Since really the 90s on, it’s been a large push to modernize music. It seems to have failed.
In our church there has been a routine of singing modern worship songs followed by one or two more traditional songs. The audience response is palpable. It’s an immediate difference.
It’s not my goal to disparage attempts at new songs but the systems in place to do it are largely not working.
I suspect in another 30-40 years we will still know the songs of the 1800-1900s far more than those of the early to mid 2000s.
Image set1 image · open to inspect
Skin in the game
It's good to have a little skin in the game of life. So I want to make some predictions.
1. Right now AI models are sufficiently new and powerful enough that the government is stepping in. We will suffer through this for another couple of years and then largely get over it. It's new, but we've seen this song and dance with internet technologies before. Intelligence will be abundant and we will adjust.
2. Local models and open-weight models will get better, more capable, and cheaper, but so will frontier. That doesn't seem to be slowing down yet, and I don't think it will for a while. This will lead to more specialized models we run for specific use cases. For example, I'd love a great local linux devsecops model that I just have run/guard my linux machines.
3. Data continues to be the moat. What you have that no one else does and what you aggregate and manage for AIs will matter more than legacy software features. For SaaS, especially, this is all that matters. Features are a commodity now.
4. We have AGI and the world hasn't ended yet. This tells us that it's less about intelligence and more about harnessing it. Right now everything else is catching up. It will happen quickly, but that's the bottleneck right now.
5. The AI boom is forcing us to invest in critical infrastructure at just the right time. We will end up benefiting more from second-order effects of abundant cheap power than we will from AI by itself. At the end of the day, power is everything to societies.
It's good to have a little skin in the game of life. So I want to make some predictions.
1. Right now AI models are sufficiently new and powerful enough that the government is stepping in. We will suffer through this for another couple of years and then largely get over it. It's new, but we've seen this song and dance with internet technologies before. Intelligence will be abundant and we will adjust.
2. Local models and open-weight models will get better, more capable, and cheaper, but so will frontier. That doesn't seem to be slowing down yet, and I don't think it will for a while. This will lead to more specialized models we run for specific use cases. For example, I'd love a great local linux devsecops model that I just have run/guard my linux machines.
3. Data continues to be the moat. What you have that no one else does and what you aggregate and manage for AIs will matter more than legacy software features. For SaaS, especially, this is all that matters. Features are a commodity now.
4. We have AGI and the world hasn't ended yet. This tells us that it's less about intelligence and more about harnessing it. Right now everything else is catching up. It will happen quickly, but that's the bottleneck right now.
5. The AI boom is forcing us to invest in critical infrastructure at just the right time. We will end up benefiting more from second-order effects of abundant cheap power than we will from AI by itself. At the end of the day, power is everything to societies.
Image set1 image · open to inspect
So let’s do a little write-up of how this web app works. Here’s a screenshot of the authoring page. You’re reading this from the notes feed page.
I have the new note option saved as a bookmark to easily pull it up and jot a note down.
I added some quality-of-life features. First is a generative AI illustration option using gpt-image-2. This lets me quickly add images to flesh out text.
Second is a tool to catch errors before they go out. This uses gpt-5.4-mini. I use it to fix spelling and basic grammar.
I also use it to add alt text automatically to images.
I have “infinite” scroll built out. There’s also RSS because I’m old school like that.
It’s all hosted on Cloudflare.
And that’s it so far! It’s been fun building it as I ran into friction or thought of new things to build in. I’m sure I’ll add more as I use it more, but for now that’s where I’ve gotten to today.
I have the new note option saved as a bookmark to easily pull it up and jot a note down.
I added some quality-of-life features. First is a generative AI illustration option using gpt-image-2. This lets me quickly add images to flesh out text.
Second is a tool to catch errors before they go out. This uses gpt-5.4-mini. I use it to fix spelling and basic grammar.
I also use it to add alt text automatically to images.
I have “infinite” scroll built out. There’s also RSS because I’m old school like that.
It’s all hosted on Cloudflare.
And that’s it so far! It’s been fun building it as I ran into friction or thought of new things to build in. I’m sure I’ll add more as I use it more, but for now that’s where I’ve gotten to today.
Image set1 image · open to inspect
I love projects like these. I honestly love logs and SIEMs. It wouldn’t hurt my feelings to work on a big SIEM product one day.
blog.nano.rs
Introducing nano | nano blog — nano / blog
nano is a lightweight SIEM in Rust on ClickHouse, with a piped query language, a real detection lifecycle, and AI that does actual investigation work. Here is what it is and why every piece is the way it is.
Thariq's insight on the usefulness of html artifacts matches my own experience with my work. I have used them a great deal lately to great effect.
x.com
Thariq (@trq212) on X
https://t.co/MXt5XS4xBX