Les Vogel

re: #DataCenters in space – I wish someone would explain to me why coronal mass ejections are no longer considered a threat? Especially as it has a 11 year rhythm? Is the lower cost of power and cooling really worth risk of a total wipeout? SpaceX lost 38 satellites in 2024 due to two CME's and his satellites were still in low earth orbit. We know that black swan events aren't really that rare and that there have been events in the last 200 years that if they were to happen today would take out most electronics on earth. Why would things that would likely be outside the ionosphere not be at much greater risk of damage?

We have been trying for decades to make it easier for less trained people to produce code. I've been programming for over 50 years. I know a lot of tricks that no one cares about today. (ex. disabling cache on a CPU during a neural net eval speeds things). I would have argued that scaling a successful project would move folks away from Vibe Coding, but we have a lot of examples where that just isn't true.

Software engineering has drastically changed in the last two years. This shift isn't about the volume of code being generated, but by the cultural shifts that came with it. Upfront design is not valued much, and it promotes a lot of prototypes to flourish and cookie licking.

@rakyll A communication & complexity tax. That’s been known since the Mythical Man Month was published 50 years ago.

More people = less progress.

@rakyll The best way to predict the future is to invent it - Alan Kay

What separates good engineers from great engineers is that great engineers can see the future (consistently).

@rakyll 10 years ago – I could keep up with launches and understood the complete product line for GCP. Now as an outsider I feel completely overwhelmed.

10 years ago, I used to have a good sense of what Google is going to launch within the next 2-3 years. Today, I barely have a sense what we’ll launch in a few months. The velocity shift is real.

@rakyll Militaries often appreciate bleeding edge technology. (They are often using things many years before consumers get access to it.)

If you want to work on something shiny and bleeding edge, work on consumer products. If you want to work on something deeply boring but stable, work on platform / cloud products.

@seanw_m @rakyll The changes to the economic system required to build these might represent things we no longer want. On the other hand, with robot servants, it might be possible.

these guys have never met a general contractor in their life

@Shruti_0810 ML

"Introduction to Machine Learning Systems" - FREE from MIT Press - Authored by Harvard Professor - 2048 Pages To Get It Simply: 1. Retweet & Reply "ML" 2. Follow so that I will DM you.

@levie Programmers and engineers have been heavy users of these things and trust is declining. newsletters.feedbinusercontent.com/bf8/bf82d1b40c… - if the folks who know the most about this stuff are being disappointed it might not be time to move trust completely for health care.

Almost every study shows doctors with AI perform better than those without. Now AI is achieving perfect scores in medical licensing exams. You will simply expect every professional services provider you go to will use AI in the future or you won’t trust the advice.

@rakyll There were a couple of candidates where I was asked not to give a coding question to and focus on system design. There were also a few managers where I was asked to pass on coding questions. I never gave SH for any of them, and noted why in the summary.

Regardless of their level, someone in the engineering ladder should be able to produce code and be able to represent their ideas in code when needed.

I'll be attending Webflow Conf in NYC - let me know if you'd like to chat.

@rakyll I'm not sure I understand the first part of this sentence.

@julesagent So, I waited over an hour and still no joy. I refreshed the page and republished and it was done in under 15s.

@julesagent It's now 10m waiting. Sad. :(

Everyone now gets 60 daily tasks powered by Gemini 2.5 Pro. During this beta period, we really want to learn how we can better support you all. Throw everything at Jules: your backlog, your meaty refactor, or try the same task from a few different approaches, etc

The Cost of Passive Avoidance feld.com/archives/2025/… @bfeld Would like to preorder your book, disappointing it's not available for Kindle.

@rakyll They do what they are trained on. Sadly, they've been trained on a lot of bad practices. Ideally, there are folks creating synthetic datasets based on good practices and training data graders to eliminate bad practices from training sets. It will take time, but we'll get there

LLMs love generating else cases 😳 It took us 30 years to teach people to avoid deep conditional nestedness and we thought we won.

@rakyll Part of that is not really understanding the motivation of customers. Capitalists ideally want to spend a bunch of money so that they can provide something that gives them "rent" forever. They donʻt really want to pay to "improve" that something forever if that donʻt have to.

Adding is easy, deleting is hard when you have real customers.

The most bizarre thing about my working for G was that it was really rare that I got to program. Now, working for myself, I get to program most days.