roadside_picnic

This seems to be the same misunderstanding about agentic coding I see a lot of places.

Agentic coding is not about creating software, it's about solving the problems we used to need software to solve directly.

The only reason I put my agentic code in a repo is so that I can version control changes. I don't have any intention of sharing that code with other people because it wouldn't be useful for them. If people want to solve a similar problem to me, they're much better of making their own solution.

I'm not at all surprised that most of Claude linked output is in low star repos. The only Claude repos I even bother sharing are those that are basically used as context-stores to help other people get up to speed faster with there of CC work.

> not a single one actually launched.

I think this represents a fundamental misunderstanding of how these AI tools are used most effectively: not to write software but to directly solve the problem you were going to solve with software.

I used to not understand this and agreed with the "where is all the shovelware" comments, but now I've realized the real shift is not from automating software creation, but replacing the need for it in the first place.

It's clear that we're still awhile away from this being really understood and exploited. People are still confusingly building webapps that aren't necessary. Here's two, somewhat related, examples I've come across (I spend a lot of time on image/video generation in my free time): A web service that automatically creates "headshots" for you, and another that will automatically create TikTok videos for you.

I have bespoke AI versions of both of these I built myself in an afternoon, running locally, creating content for prices that simply can't be matched by anyone trying to build a SaaS company out of these ideas.

What people are thinking: "I know, I can use AI to build a SaaS startup the sells content!" But building a SaaS company still requires real software since it has to scale to multiple users and use cases. What more and more people are realizing is "I can created the content for basically free on my desktop, now I need to figure out how to leverage that content". I still haven't cracked the code for creating a rockstar TikTok channel, but it's not because I'm blocked on the content end.

Similarly I'm starting to see that we're still not thinking about how to reorganize software teams to maximally exploit AI. Right now I see lots of engineers doing software the old way with an AI powered exo-skeleton. We know what this results in: massive PRs that clog up the whole process, and small bugs that creep up later. So long as we divide labor into hyper focused roles, this will persist. What I'm increasingly seeing is that to leverage AI properly we need to re-think how these roles actually work, since now one person can be much responsible for a much larger surface area rather than just doing one thing (arguably) faster.

This was my thought as well. Trump today claims he had long discussions with Iranian leadership over the weekend and Iran claims such a thing never happened.

Not only is Iran justifiably angry at its attackers, they have no reason to trust a ceasefire would be respected, and, furthermore, despite damage taken, stand to gain quite a lot if the war progresses as it has. Currently Iran is exporting more oil than ever before[0]. If the war continues as is they stand to emerge a much, much more dominant power on the world stage. Essentially being to BRICS what the GCC countries have been to the US (although much better defended).

Even the threats against their energy infrastructure can't really be backed up effectively. Iran has intentionally built its energy infrastructure to be highly distributed. Meanwhile Israel's infrastructure is highly centralized, with a small number of power plants providing massive proportions of the nations infrastructure. All this meaning that Iran can retaliate much more effectively the they can be harmed.

0. https://www.wsj.com/world/middle-east/irans-control-of-hormu...

That more-or-less happened over the weekend [0]. Five minutes before Trump's unsurprising TACO announcement an unexpected large number of WTI futures contracts were sold and an even larger amount of S&P 500 futures were purchased.

0. https://www.cnbc.com/amp/2026/03/23/volume-in-stock-and-oil-...

> like entity recognition

As someone who has done traditional NLP work as at least part of my job for the last 15 years, LLMs do ofter a vastly superior NER solution over any previous NLP options.

I agree with your overall statement, that frequently people rush to grab an LLM when superior options already exist (classification is a big example, especially when the power of embeddings can be leveraged), but NER is absolutely a case where LLMs are the superior option (unless you have latency/cost requirements to force you to choose and inferior quality as the trade off, but your default should be an LLM today).

I'm a pretty big fan of Roadside Picnic!

> develop new technology, research culture solutions.

The technology and culture solutions have existed and been evolving for 20 years. It really sounds like your experience with remote work is not representative.

Sounds like you don't have a lot of remote work experience.

The majority of my career (years before the pandemic) has been remote work. I find in office work painfully slow. I pair program quite often remote, and when someone gets stuck we also "just talk". Honestly I prefer screen sharing to leaning over someone's shoulder (much easier to doing supporting work in parallel).

I find it really depends on the type of org though. Large corporate places do tend to suffer from remote work because so much of the work is performative anyway. Remote small companies and startups the velocity is very high, but you do need more senior people capable of independent work.

Especially when you factor in the easy of "after hours" work, the amount of emergency stuff I've shipped around midnight is incomparable to the 'in office' equivalent.

Though I suspect the key word here is "my 2 programmers", I find managers don't feel like their doing work unless they're physically watching it get done.

Not understanding how to run a remote team is not the same as remote teams not being effective in principle.

> The use of the word agents is interesting is mostly a coincidence, it is used today in a sense that didn't quite exist 2 years ago

I'm sorry but it's wild to me that you could write so much about "agents" without recognizing their long, established history in computer science (especially in AI) outside of OOP. Agents are basically the entire framing of Norvig and Russel's "Artificial Intelligence: A Modern Approach" [0] (originally published in 1995, but drawing from much earlier work).

Not specifically AI, but not unrelated either, Agents play a major role in how we understand concurrency and mobile communication. The author of this paper, Robin Milner, is responsible, among many other things, for establishing the π-calculus (1992), which defines a formal language to describe agent communication.

If you want to go closer to the source you can take a look at Hewett's "Actor Model" [2] 1973. Which is when the field first started to formalize the idea of software agents.

The current use of the word "agent" is basically a marketing buzz-word that largely ignores the decades of research in the field of computer science around how to design intelligent interacting agents to accomplish tasks. Which is a bit of a tragedy because I personally think current LLMs could gain a lot of value if thought about in the traditional agent sense.

0. https://people.engr.tamu.edu/guni/csce625/slides/AI.pdf

1. https://www.pure.ed.ac.uk/ws/files/16426053/A_Calculus_of_Mo...

2. https://www.ijcai.org/Proceedings/73/Papers/027B.pdf

We will never recovered from the fact that "literally" now means "figuratively but with emphasis".

The apprenticeship was after this was painted. Prior to engaging in formal training he largely ignored school and spend his time painting and seeking out other painters to learn from:

> As a young boy, Michelangelo was sent to the city of Florence to study grammar under the Humanist Francesco da Urbino. Michelangelo showed no interest in his schooling, preferring to copy paintings from churches and seek the company of other painters. [0]

You can search other sources and you'll find the same: prior to apprenticeship he was not formally enrolled in any form of art education.

0. https://en.wikipedia.org/wiki/Michelangelo

Many of the comments here are expressing disbelief that this could have been created by a 12 year old, but people fail to recognize that, not only did Michelangelo have tremendous natural talent, but grow up in a world where, as a child, he was allowed to spend enormous amounts of his time and energy studying with professional artists.

He wasn't being dropped off a school at 7am, squirming in a chair until 3pm, playing video games before dinner and then doing homework until bed all while squeezing in a bit of time for sketching.

The vast majority of people probably benefit more from our current structure, but it does make it much less likely to have "genius" of the type we see in Michelangelo, Mozart, etc.

Thankfully social media is getting so much worse so fast it's making this easier and easier. HN is the last social media platform I still participate in... and I suspect that might not be for too much longer.

I recently logged onto Facebook and Instagram to update my 2-factor auth settings after having too many notifications of malicious login attempts. It was incredible to see what a transformation has happened there, it's like going to a decaying suburban shopping mall with only a few stores left open (and sort of sad to see the remaining users so continually desperate for a drop of approval from some imagined community).

Reddit is mostly bots, astro-turfers and people so brainwashed it's hard to tell the difference. I remember disagreeing with people on there (this in the pre-Digg migration era) you would get interesting divergent points of view. Now it's like people are reading from a script.

Twitter used to be my strongest addiction, but it's almost unbelievable how big a transformation has occurred since it became X. It's almost a parody of everyone's dystopian social media fears.

HN has obviously held up a bit better, but the AI driven mass hallucination impacting this community, combined with the increasingly aggressive manipulation of the home page, is continually making logging out for good seem like the best option.

I can't tell if this is a tongue-in-cheek comment or not, but all of that is shipped off to 3rd party "recyclers" who pinky promise that they will dispose of it properly. Very often those 3rd parties rely on other 3rd parties until the it ends up in a waste pile in a developing country, but with a long enough chain of differed responsibility that nobody can be held accountable.

The fundamental problem with "recycling" is precisely the fact that we just hand it off and don't ask questions about where it ends up, all while feeling great about ourselves afterwards. Bestbuy and Staples are offering accountability laundering so that you don't have to feel bad and in exchange are more likely to become a customer. The 3rd parties working for them do the same thing, but they usually want cash for it.

> "it's a scam, just chuck it all in the garbage"

This sentiment is the case because very often that's where recycling ultimately ends, we just pay someone to move it far away from us so we don't have to see it when it happens.

Until 2018, when they finally stopped accepting it, one of the US largest exports to China was cardboard boxes sent over for "recycling". We burned tons of bunker fuel shipping back the boxes Chinese goods arrived in. The net environmental impact would likely have been less had we just kept the boxes at home.

It's strange to me how often people prefer a widely acknowledged lie than to simply admit the truth.

I always recycle though because the recycle bin in my city is larger than my trash bin, and I don't have enough room in my trash bin sometimes.

> It’s beautiful

Especially since both the waste created in the process of making the device and the e-waste created with it's disposal are somebody else's problem!

> young founders

While this place has always been attractive to people building startups, back in the day (my original account is from 2009) "Hacker" News was much more about Hackers. Most people posting here had read "On Lisp", respected Paul Graham as a programmer and were enthusiastic about programming and solving problems above all else.

I'm honestly curious how many people that visit HN today even know what a "y combinator" is, and I have a pretty reasonable guess as to how many have implemented it for fun (though probably the applicative order version).

> Because we could buy stuff without leaving the house?

I'm guessing you were still pretty young/not yet born at the time?

Online shopping didn't just mean "I don't have to leave the house", it opened up a whole world of what was even possible.

Prior to the web if you didn't live in a big city (and less people did then) then your access to books, music, movies was insanely restricted.

I deeply recall how painfully limited my local Sam Goody was, even major "alternative" bands only had partial discographies available. I remember visiting my father's college campus as an early teen and being beyond excited to find a copy of NIN's "halo 1" in a college town record store. True indie music was reserved for kids with cool older siblings that both knew where the stores were and could drive there. In order to watch Dragon Ball Z I had to rely on a friend whose dad was a plumber in NYC and knew where the bootleg stores in China-town were. I got to tag along once and picked up a single random episode of a Gundam series, never to be finished because I could never find another.

Sure it was sort of fun to figure all this stuff out, but at the same time my bookshelf is filled with books that changed my life in various ways I would simply never had been able to find (or even be aware of) in the pre-web days. If you wanted to learn programming in the 90s you had to hope your local Walden books had some good options, and you certainly weren't going to learn Haskell or Lisp. Mine only had books on Excel, so I didn't learn to program until I was older.

Now the fact that American suburbs where a complete cultural wasteland in the 90s might be the bigger issue than the cure which was the web, but nonetheless the early web did make the world of information much bigger.

I'm not a particularly big Gary Marcus fan, but I'm even less of a fan of the term "always".

I browsed back looking for posts of his that most obviously made predictions, and "GPT-5… now arriving Gate 8, Gate 9, Gate 10" make a few very clear predictions and was absolutely correct [0] about them.

This was in June 2024, and Marcus claims two major predictions:

- "As of today, I am more confident than ever that GPT-5 won’t land this year."

- "Gary Marcus is still betting that GPT-5 will continue to hallucinate and make a bunch of wacky errors, whenever it finally drops."

It would be over a year from his posting that article that GPT-5 would finally land, and his overall prediction that the result would be lack-luster was also spot on.

Again, I don't particularly care one way or another about Gary Marcus, but flat out dismissing his writing doesn't really hold up.

0. https://garymarcus.substack.com/p/gpt-5-now-arriving-gate-8-...

Thank you, I had corrected it earlier when I had some time to further investigate what was happening.

Formal verification combined with AI is, imho, exactly the type of thinking that gets the most value out of the current state of LLMs.

You are correct, I assumed context was implied, but I do mean "recent work with LLMs". Friends of mine where doing side projects with him about two decades ago, and I have a few of his books on my shelf, so yes, I am aware that Terry Tao was doing work in mathematics prior to the advent of LLMs.

> It is still likely very tedious to confirm what the LLMs say,

A large amount of Tao's work is around using AI to assist in creating Lean proofs.

I'm generally on the more skeptical side of things regarding LLMs and grand visions, but assisting in the creation of Lean proofs is a huge area of opportunity for LLMs and really could change mathematics in fundamental ways.

One naive belief many people have is that proofs should be "intelligible" but it's increasingly clear this is not the case. We have proofs that are gigabytes (I believe even terabytes in some cases) in size, but we know they are correct because they check in Lean.

This particular pattern of using state of the art work in two different areas (LLMs and theorem proving) absolutely has the potentially to fundamentally change how mathematics is done. There's a great picture on pp 381 of Type Theory and Formal Proof where you can easily see how LLMs can be placed in two of the most tricky parts of that diagram to solve.

Because the work is formally verified we can throw out entire classes of LLM problems (like hallucinations).

Personally I think strongly typed language, with powerful type systems are also the long term ideal coding with LLMs (but I'm less optimistic about devs following this path).

Projects are essential, but I've found there is a huge problem with your advice: you have no clue about the possible solution surface.

My advice to learners has been "try to learn as much about a topic as someone who has taken the subject in college and forgotten about it".

For example consider calculus: Someone who took calc 20 years ago and hasn't used it since will probably forget exactly how to compute most derivatives and integrals. But if someone mentions an optimization problem "we need to know when this curve peaks" or asks something involving finding the area under a curve, alarm bells should start ringing. They'll know this can be done, and likely go grab a calc book to refresh.

Another example I run across all the time, which is the opposite scenario: Survival analysis. I have been on countless teams where somebody needs to understand something like churn or the impact of a offering a discount that hasn't expired yet, etc. These are all classic survival analysis problems, yet most people are ignorant that this field of study even exists! Because of this I've seen so many times where people complain that "we'll have to wait months or years to see if these changes impact customer lifetime!" (note: if anyone out there is doing Churn or LTV analysis and aren't familiar with survival analysis, you are most certainly approaching it incorrectly).

I've seen a lot of people get frustrated with self study because they try to learn the material too well. If you aren't going to be using survival analysis soon, it's not really worth remembering all the details of how to implement a Kaplan Meier curve. But if you even have a vague sense of what problem this solves, when you encounter that problem in a project, you know where to go back to. Then you typically walk away with a much stronger sense of the subject then if you had studied it harder in the first place.

> negligible odds of death or assault?

Negligible odds of death I'll buy, but a very large number of my women friends over my life have been sexually assaulted (and probably far more than I realize, it's not like it's something you bring up during holiday dinner). I'm often shocked by how few men realize how prevalent this is.

The idea that "women who aren't comfortably traveling the world alone depending on the kindness of strangers are living a life ruled by fear" seems naive at best.

Even back in the days when hitchhiking was much more common you would almost never see a single woman by herself, for good reason.

> a wave of elective sociopaths swoop in to flag posts and derail discussion

The biggest vulnerably of the entire social media model is the "engagement pyramid" where the number of viewers is much larger than the number of upvoters is much larger than the number of commenters.

HN gets ~5 million monthly unique visitors, or 150,000 per day (conservatively assuming each visitor only shows up once, so that number is probably much larger). But if you look the top post right now has around ~1,500 upvotes, and ~200 comments (and, if you look, comment sections frequently have single users commenting repeatedly, so less than 200 actual people commenting).

This makes it very easy and very worth it to run even loosely coordinated commenting/upvoting rings. 10 people can easily downvote a new post off the homepage, give the impression that the community disapproves of a certain opinion, disrupt conversation, etc.

What's funny is that proposing this is often treated as a claim of a wild conspiracy theory. But the weakness in most conspiracy theories is that they require high levels of coordination among similarly large groups of actors, often for little reward. In this case it's almost more outrageous to claim that this isn't happening (especially since I have personally seen teams of people coordinate to get their startups work on the front page). I suspect even 3 very coordinated people could do a lot to control the front page (or alternatively increasingly large N of increasingly less coordinated people).

> One might as well say that neural networks trained with gradient descent are a reinvention of numerical methods for function approximation.

I don't know anyone who would disagree with that statement, and this is the standard framing I've encountered in nearly all neural network literature and courses. If you read any of the classic gradient based papers they fundamentally assume this position. Just take a quick read of "A Theoretical Framework for Back-Propagation (LeCun, 1988)" [0], here's a quote from the abstract:

> We present a mathematical framework for studying back-propagation based on the Lagrangian formalism. In this framework, inspired by optimal control theory, back-propagation is formulated as an optimization problem with nonlinear constraints.

There's no way you can read that a not recognize that you're reading a paper on numerical methods for function approximation.

The issue is that Vaswani, et al never mentions this relationship.

0. http://yann.lecun.com/exdb/publis/pdf/lecun-88.pdf

I will beat loudly on the "Attention is a reinvention of Kernel Smoothing" drum until it is common knowledge. It looks like Cosma Schalizi's fantastic website is down for now, so here's a archive link to his essential reading on this topic [0].

If you're interested in machine learning at all and not very strong regarding kernel methods I highly recommending taking a deep dive. Such a huge amount of ML can be framed through the lens of kernel methods (and things like Gaussian Processes will become much easier to understand).

0. https://web.archive.org/web/20250820184917/http://bactra.org...

The key word here is "Wall Street". And this statement is playing off a popular misconception around corporate investors buying up American houses.

There has been a bit of a panic around "Investors buying up all the property!!!" With people often citing Black Rock and Blackstone as the main culprits. But most of the "investors" buying up property are individuals purchasing investment properties.

Here's an article on the topic from 2023[0], a bit old but my understanding is large institutional investment in residential real estate was already starting to cool down.

Black rock isn't buying up all the housing, your neighbors are.

I suspect this statement, and even if it becomes an actual ban, is largely to gain wider popular support around a largely imaginary concern people have.

0. https://www.housingwire.com/articles/no-wall-street-investor...

I've lived through multiple incredibly rapid changes in tech throughout my career, and the lesson always learned was there is a lot of wasted energy keeping up.

Two big examples:

- Period from early mvc JavaScript frontends (backbone.js etc) and the time of the great React/Angular wars. I completely stepped out of the webdev space during that time period.

- The rapid expansion of Deep Learning frameworks where I did try to keep up (shipped some Lua torch packages and made minor contributions to Pylearn2).

In the first case, missing 5 years of front-end wars had zero impact. After not doing webdev work at all for 5-years I was tasked with shipping a React app. It took me a week to catch up, and everything was deployed in roughly the same time as someone would have had they spent years keeping up with changes.

In the second case, where I did keep up with many of the developing deep learning frameworks, it didn't really confer any advantage. Coworkers who I worked with who started with Pytorch fresh out of school were just as proficient, if not more so, with building models. Spending energy keeping up offered no value other than feeling "current" at the time.

Can you give me a counter example of where keeping up with a rapidly changing area that's unstable has conferred a benefit to you? Most of FOMO is really just fear. Again, unless you're trying to sell your self specifically as a consultant on the bleeding edge, there's no reason to keep up with all these changes (other than finding it fun).

I feel similar, I'm not against the idea that maybe LLMs have gotten so much better... but I've been told this probably 10 times in the last few years working with AI daily.

The funny part about rapidly changing industries is that, despite the fomo, there's honestly not any reward to keeping up unless you want to be a consultant. Otherwise, wait and see what sticks. If this summer people are still citing the Opus 4.5 was a game changing moment and have solid, repeatable workflows, then I'll happily change up my workflow.

Someone could walk into the LLM space today and wouldn't be significantly at a loss for not having paid attention to anything that had happened in the last 4 years other than learning what has stuck since then.