My experience was that enterprise programmers burned out on things like WSDL at about the same time Rails became usable (or Django if you’re that way inclined). Rails had an excellent story for validating models which formed the basis for everything that followed, even in languages with static types - ASP.NET MVC was an attempt to win Rails programmers back without feeling too enterprisey. So you had these very convenient, very frameworky solutions that maybe looked like you were leaning on the type system but really it was all just reflection. That became the standard in every language, and nobody needed to remember “parse don’t validate” because heavy frameworks did the work. And why not? Very few error or result types in fancy typed languages are actually suited for showing multiple (internationalised) validation errors on a web page.
The bitter lesson of programming languages is that whatever clever, fast, safe, low-level features a language has, someone will come along and create a more productive framework in a much worse language.
Note, this framework - perhaps the very last one - is now ‘AI’.
The most interesting thing about everyone trying to position themselves as AI experts is the futility of it: the technology explicitly promises tomorrows models will be better then todays, which means the skill investment is deflationary: the best time to learn anything is tomorrow when a better model will be better at doing the same work - because you don't need to be (conversely if you're not good at debugging and reverse engineering now...)
the best time to learn anything is tomorrow when a better model will be better at doing the same work
doesn’t that presume no value is being delivered by current models?
I can understand applying this logic to building a startup that solves today’s ai shortcomings… but value delivered today is still valuable even if it becomes more effective tomorrow.
I think it also presumes that the skills of today won't be helpful in making you better, faster, stronger at knowing what to learn tomorrow. Skateboarding ain't snowboarding but I guarantee the experience helps.
I'm pretty much just rawdogging Claude Code and opencode and I haven't bothered setting up skills or MCPs except for one that talks to Jira and Confluence. I just don't see the point when I'm perfectly capable of writing a detailed prompt with all my expectations.
The problem is that so many of these things are AI instructing AI and my trust rating for vibe coded tools is zero. It's become a point of pride for the human to be taken out of the loop, and the one thing that isn't recorded is the transcript that produced the slop.
I mean, you have the creator of openclaw saying he doesn't read code at all, he just generates it. That is not software engineering or development, it's brogrammer trash.
I think the rationale is that with the right tools you can move much faster, and not burn everything to the ground, than just rawdogging Claude. If you haven't bothered setting up extra tools you may still be faster / better than old you, but not better than the you that could be. I'm not preaching, that's just the idea.
> That is not software engineering or development, it's brogrammer trash.
Yes, but it's working. I'm still reading the code and calling out specific issues to Claude, but it's less and less.
That’s true for “tips and tricks” knowledge like “which model is best today” or “tell the model you’ll get fired if the answer is wrong to increase accuracy” that pops up on Twitter/X. It’s fleeting, makes people feel like “experts”, and doesn’t age well.
On the other hand, deeply understanding how models work and where they fall short, how to set up, organize, and maintain context, and which tools and workflows support that tends to last much longer. When something like the “Ralph loop” blows up on social media (and dies just as fast), the interesting question is: what problem was it trying to solve, and how did it do it differently from alternatives? Thinking through those problems is like training a muscle, and that muscle stays useful even as the underlying technology evolves.
It does seem like things are moving very quickly even deeper than what you are saying. Less than a year ago langchain, model fine tuning and RAG were the cutting edge and the “thing to do”.
Now because of models improving, context sizes getting bigger, and commercial offerings improving I hardly hear about them.
> what problem was it trying to solve, and how did it do it differently from alternatives?
Sounds to me like accidental complexity. The essential problem is to write good code for the computer to do it's task?
There's an issue if you're (general you) more focused on fixing the tool than on the primary problem, especially when you don't know if the tool is even suitable,
You nailed it. Thats exactly how I feel. Wake me up when the dust settles, and i'll deep dive and learn all the ins and outs. The churn is just too exhausting.
I don't get the pressure. I don't know about you, but my job for a long time has been continually learning new systems. I don't get how so many of my peers fall into this head trip where they think they are gonna get left behind by what amounts to anticipated new features from some SaaS one day.
How do you both hold that the technology is so revolutionary because of its productive gains, but at the same time so esoteric that you better be ontop of everything all the time?
This stuff is all like a weird toy compared to other things I have taken the time to learn in my career, the sense of expertise people claim at all comes off to me like a guy who knows the Taco Bell secret menu, or the best set of coupons to use at Target. Its the opposite of intimidating!
I may just be a "doomer", but my current take is we have maybe 3-5 years of decent compensation left to "extract" from our profession. Being an AI "expert" will likely extend that range slightly, but at the cost of being one of the "traitors" that helps build your own replacement (but it will happen with or without you).
> the technology explicitly promises tomorrows models will be better then todays, which means the skill investment is deflationary
This is just wrong. A) It doesn’t promise improvement B) Even if it does improve, that doesn’t say anything about skill investment. Maybe its improvements amplify human skill just as they have so far.
I have a reading list of a bunch of papers i didn't get through over the past 2 years. it is crazy how many papers on this list are completely not talked about anymore.
I kinda regret going through the SeLU paper lol back in the late 2010s.
Either the business makes a profit before it gets swept away, or it doesn't. This should be your goal: make money before your business dies. If you do that, you succeeded. Businesses are always ephemeral.
This is very well put. I think this platform can be useful but I doubt it can be something as big as the think it will be. At the end of the day it’s just storing some info with your data. I guess they are trying to be the next GitHub (and clearly have the experience :)). I doubt that success can be replicated today with this idea, even with $60 mil to burn
Hudl is the leader in sports technology, video analysis & data. We're looking for the best talent around the world. Come help teams and athletes reach their potential—and find yours, too.
Offices in the US, Netherlands, India, UK, Italy, China, Japan, Spain, New Zealand, Australia. Multiple roles across hardware, IT, MLOps, engineering, sales and beyond:
Juggling is one of hobbies with the highest ratios of being able to impress random people versus the actual effort you have to put in, and I generally find I never forget 3-ball stuff I learned as a kid. It's also as good as a long walk for getting you out of your head when needed.
Shout out to anyone that remembers the Mushy Pea juggling shop in Manchester many years ago, where I learned all sorts of circus skills.
Rubik's cubing is another. Most people here with enough logical aptitude to be programmers could probably learn a beginner method in a day or two. I'll pick up a cube anywhere I see one scrambled, solve it in a couple minutes, and then the last flourish is to leave it in the checkerboard pattern.
I also juggle, and the result of the combination is that approximately every single person on Facebook has posted to me the video about solving cubes while juggling them...
The biggest difference about Rubik's cubing is that once you have mastered 3x3x3, 4x4x4, and 5x5x5 (which is just a slight variant on 3x3x3) all subsequent cubes are just natural extensions of these. A 17x17x17 requires no new skills that aren't already mastered in a 5x5x5, it just takes more time. The nice part of about this is you can really blow peoples minds by pulling out an 11x11x11 cube and solving it at though it were nothing.
Even sided cubes are the hardest because they have issues with "parity" that are only uncovered near the end of the solve and can be quite tedious to fix (at least imho).
Right, cubing scales up much more readily than juggling. Although I've found the bigger cubes aren't quite so impressive because you can't do them as fast. I take 10-15 minutes for the 5x5x5, and by then any observer has lost interest.
But yeah, some people will be impressed just because it's bigger. They always say that it must be super hard. My stock reply: "It's like a jigsaw puzzle. Is 1000 piece harder than 500? Not really harder, it's just more of the same thing." Sometimes that gets a blank look, sometimes that induces enlightenment.
My favorite cuboid variant is a 3x3x4. That blows people's minds when they see it's not symmetric, and they handle it and realize that it can't do certain turns (the long axes have to be 180° not 90°), but in fact that limitation makes it easier and I can solve it almost as fast as a 3x3x3.
Juggling 3 is a skill that is way easier than people think until they do it. But the very next question is will be, "how many can you juggle?" as they apparently think juggling 4 is just 33% more difficult than 3.
It’s also fun that you can tweak it without the layperson even noticing the change in relative difficulty.
Cascade pattern = easy difficulty
Shower pattern = normal difficulty
Box pattern = hard difficulty
As someone who loves to run their hands up and down in the piano in grand sweeping arpeggios, I'm a huge fan of patterns where the perceived difficulty is higher than the actual difficulty.
For those wondering: to juggle 4 balls, you either have to decrease the time between catching a ball and throwing it again or increase the time a ball is in the air.
Unless you start throwing feathers or balloons, the latter requires you throw higher. That requires you to either spend more time launching them up (bad for the ‘decrease time between catching and throwing’) or use more force (bad for throwing accuracy.
Also, even assuming you juggle 4 balls keeping “time in hand” equal, you have to throw it higher by a factor of (4/3)². That’s almost 2.
And even if you manage to make those throws with the same accuracy in angles, the errors in location by the time you catch the balls scale by the same factor.
> you either have to decrease the time between catching a ball and throwing it again or increase the time a ball is in the air
I think you might be thinking of 5 ball juggling.
4 ball juggling (or at least it's most common variant, "The Fountain" [0]) is fascinating because it's really juggling two balls in each hand in a way that makes it appear similar to the standard cascade. Though this may sound "less hard" than what people initially imagine, it's a very different feeling than all the basics you learn using only 3 balls.
They're not wrong. Assuming alternating hands are doing the throwing (rather than both hands throwing a ball simultaneously), 4-ball patterns mean each ball is in the air for 4 beats, while 3-ball patterns only take 3 beats each.
And once one realises that many juggling patterns can be understood by the number of beats each ball takes to return to a hand, one can then think in siteswap (https://juggle.fandom.com/wiki/Siteswap).
In my experience people can’t even tell how many balls you’re juggling after 3. 4, 5, 6, 7 all get “how many balls is that”, silly stuff like 3 ball factory blows more minds than 5 ball site swaps
Also, with a three ball pattern, most of the time there's only a single ball in the air. With four, there are almost always two. The odds of a mid-air collision increase significantly, and go up as the number of balls increases.
That also forces you to have much more consistent throws which, as you note, gets harder because you also have to throw higher which scales up any error in the force you're applying.
I've also found it to be a magical incantation to silence crying babies. Sometimes I'll quickly flash (juggle for one round) three random objects to shut up a baby in public and their parents don't even notice.
You need the model to interpret documentation as policy you care about (in which case it will pay attention) rather than as something it can look up if it doesn’t know something (which it will never admit). It helps to really internalise the personality of LLMs as wildly overconfident but utterly obsequious.
This is a minor thing, but as an introvert, I really try and push myself to model social behaviour to my kids. Saying good morning to people in the street, chatting to other dog owners, being nice to waiters, travelling by bus, there are lots of tiny opportunities every day to show that world is full of lovely people who aren’t scary at all.
I installed Bazzite on a slightly esoteric machine, a 16" dual-screen Asus laptop. It's not really my cup of tea as a distribution, either philosophically or practically, but it has some specific patches for Asus hardware and as a result it seems to work better in every way than Windows. Every couple of months I'm annoyed for a moment by all the immutability stuff and the package system, but for both work and play it's running perfectly.
reply