HNHacker News
TopNewBestAskShowJobs

robot-wrangler

639 karma · joined September 10, 2025

submissionscomments
robot-wrangler··on Short Little Difficult Books
> Extremely difficult for no reason whatsoever. Terrible book.

Many people believe this is the mythical "Great American Novel" we've been arguing about and/or anticipating forever. Strong argument for that because it's actually very Hollywood, isn't it? It's an absurdly action-packed cowboy-horror mashup that's full of gratuitous violence and manifest destiny.

I know several people that thought it was "extremely gory for no reason whatsoever" and did not finish for that reason, but none that thought it was difficult. I was surprised TFA mentioned it in that light, because I remember Child of God being more of a slog and BM being a page-turner.

https://en.wikipedia.org/wiki/Great_American_Novel

robot-wrangler··on Short Little Difficult Books
Agreed, good list for HN. Especially Calvino, Cormac, and PKD tend to be popular with math nerds. Also if PKD qualifies as difficult enough, maybe Borges can fit here too.

Not fiction or difficult, but I suggest Calvino's "Six Memos for the Next Millennium". It's novella-length lecture series for writers/readers, but I'd say it's really a critical/creative framework that's as relevant for math as it is for myth, or novels, or code.

robot-wrangler··on Short Little Difficult Books
> why would you want to suffer and anyone who would read such a book must be pretentious, phony brodernist snob!

Ok, this nerd-sniped me. Are we saying this now? Feels like a reference. Found it: https://lareviewofbooks.org/article/against-high-brodernism/

WTF? If the term is about superficial depth wouldn't the bros be into like, abridged graphic-novel adaptations of difficult Russian classics, and not actually diving into genuinely difficult stuff regardless of quality? Shouldn't highbrow literary types sneer with greater cleverness or at least clarity?

> I also find it strange to even worry about “pretentious” readers or “brodernists” hyping themselves up to read some new, huge tome (Schattenfroh this season it seems) in an age when few people read anything at all. Reading a very long book always takes some dedication, some challenging of yourself. So what?

Ok good, TFA actually does get it

robot-wrangler··on Nearly all UK drivers say headlights are too bright
> This is exactly what high beams are for,

Judging from comments in this thread, large numbers of people are suggesting that they are actually entitled to blind others because toggling back and forth is an inconvenience for them, and/or the "smart" cars that are auto-toggling high beams have left many drivers completely ignorant that toggling is actually possible.

Related question, are cars that have completely removed manually controlling high beams actually street legal?

robot-wrangler··on Nearly all UK drivers say headlights are too bright
> I understand that currently this is sort of a collective action problem,

There's something really obnoxious and antisocial about this that makes me really mad. Test your lights on yourself, people. If your manufacturer did something stupid here it's unfortunately on you to fix it. Usually the mentality of "got mine, fuck everyone else" is self-destructive but maybe not obviously so because cause/effect are a few steps away from each other. But if you blind everyone else on the road so you can see better then it's kind of endangering you directly and immediately!

This is a huge issue in US western states especially, since they are full of long dark drives. You'll literally be blinded for several seconds if you encounter another car even if you're averting your eyes. Bad but not horrible if it's 1/10 of chance encounters that are antisocial, but it's been getting worse for years and odds are now much closer to 1/2.

robot-wrangler··on The Miracle of Wörgl
IANAE, but besides things like the threatened oligarchs and taxes and government-monopoly on the legal use of force.. raw materials and imports/exports would seem to be an even more basic problem for an optimist that thinks the first big 3 issues can be worked around. That's why I think the dutch circular economy[1] is so interesting. Hopefully someone much more informed could opine?

It's definitely got a "degrowth" aspect involved and the captains of industry will not appreciate it. On the other hand, it's much less threatening to entrenched interests compared to a large group of people who are explicitly trying to just opt out of a system you're deeply invested in.

[1] https://www.government.nl/topics/circular-economy/circular-d...

robot-wrangler··on What if you don't need MCP at all?
MCP is how you wrap/distribute/compose things related to tool-use. Tool-use is how you insist on an IO schema that LLMs must conform to. Schemas are how you combat hallucination, and how you can use AI in structured ways for things that it wasn't explicitly trained on. And this is really just scratching the surface of what MCP is for.

You can throw all that away by rejecting MCP completely or by boiling tool-use down to just generating and running unstructured shell commands. But setting aside security issues or why you'd want to embrace more opportunities for hallucination instead of less.. shelling out for everything is perfect faith in the model's ability to generate correct bash for an infinite space of CLI surfaces. You've lost the ability to ever pivot to smaller/cheaper/local models, and now you're more addicted to external vendors/SOTA models.

Consider the following workflow with a large CLI surface that's a candidate for a dedicated LLM tool, maybe ffmpeg. Convert the man page to a JSON schema. Convert the JSON schema to a tool. Add the tool to a MCP server, alongside similar wizards for imagemagick/blender. The first steps can use SOTA models if necessary, but the later steps can all feasibly work for free, as a stand-alone app that has no cloud footprint and no subscription fee. This still works if ffmpeg/blender/imagemagick were private custom tools instead of well-known tools that are decades old. You can test the tools in offline isolation too. And since things like fastmcp support server composition you can push and pop that particular stack of wizards in or out of LLM capabilities.

Good luck getting real composition with markdown files and tweaking prompts for tone by adding a "Please" preamble. Good luck engineering real systems with vague beliefs about magic, no concrete specifications for any part of any step, constantly changing external dependencies, and perfect faith in vendors.

robot-wrangler··on America Is All-In on Deep Learning; China Emphasises Robotics and Hardware
> [China strategy] is based [..] on a few pillars: Embodied AI—robotics, advanced sensors, AI-enabled hardware; Fast-following in AI, especially with open-source models that blunt the impact [..] Adoption of AI in the here and now—building scaffolding, data pipelines, and other tweaks to make models work in businesses, and especially factories.

So in summary, a practical and realistic strategy aligned with the real world, involving consulting and cooperating with experts of many stripes, facing the world as it actually is, assessing things instead of assuming you always get what you want, planning and hedging around the uncertainty you cannot remove, doing something that helps sooner rather than later. Making things, doing things, it's just grounded

> The U.S. economy is increasingly a highly leveraged bet on deep learning. [..] Of all the bets to stake one’s economy on, deep learning is a very good one. Sam Altman’s mantra is true: deep learning works.

So in summary, an economic/strategic trajectory that's aligned with the mentality of politicians and MBAs. Which is to say.. a posture that's more or less like a child who insists on imagining they always get what they want, usually includes reality-denial for tactical purposes of lying and grifting, is out of touch with the world when they do pay attention to it. The issue with this is that these people always assume everything is about email-jobs that require no actual expertise in anything, because that's what everyone they interact with does. Do nothing but spin, make nothing but deals, act unilaterally without consulting anybody, and probably systematically just remove any experts they have access to since truth threatens power.

robot-wrangler··on Structured outputs on the Claude Developer Platform
https://github.com/pydantic/pydantic-ai/issues/582 https://github.com/pydantic/pydantic-ai/issues/2405
robot-wrangler··on The terminal of the future
Emacs user of a highly customized and well-loved setup for over a decade before I gave up the habit ;) But this illustrates my point perfectly. That's a huge list of stuff that all needs to be turned on or configured in various ways. Some newbie who is shopping for a new terminal-emulator saw this, gave up immediately, and decided to download kitty because it looks like an app and not a platform.

To successfully argue that it's just perfect as a terminal emulator, I think you need to find a way to ship it in exactly that configuration. That would mean that you open it up to a shell prompt with a dollar sign, you can hit ctrl-t to get a new terminal tab. Clicking URLs should open them in a browser without having to copy/paste. Speaking of copy/paste, that should work too, and ctrl-e, and ctrl-a, etc, etc.

robot-wrangler··on The terminal of the future
For raw data type of applications, it's definitely important to be able to preserve pipeline oriented use-cases like `cat img.png | convert-stdin-to-jpg | convert-stdin-back-to-png | imgcat`. But in the hypothetical world where all CLI I/O moves towards JSON, I'd still argue this is not only possible but now strictly easier because you can just explicitly embed mimetype info + b64 instead of assuming/detecting encoding or requiring user to specify it, or working with heuristic file-magic. (Jupyter notebooks work like this to persist images instead of just flat text.) And presumably a smarter suite of JSON oriented tools like jcat/jless would either actually display that data, or notice the type and avoid spamming the screen with raw b64.
robot-wrangler··on The terminal of the future
I guess I don't see those as big downsides because I don't think people usually want binary data or quoted strings back from a CLI command, nor do they want column oriented output, nor "user friendly" tables.

Answering --help with JSON is a good example, how bad is it really if the response is JSON? Well, using less works fine still and you can still grep if you want simple substring search. Wanting a section is probably more common, so maybe you'd "grep" for a subcommand with `jq .subcommand` or an option with `jq .subcommand.option`, and maybe get yourself a fancier, JSON-friendly version of less that handles escaped quotes and newlines. Tables and tab-or-space delimited output overflow char limits, force the command-generator to figure out character wrapping, and so on. Now you need a library to generate CLI help properly, but if you're going to have a library why not just spit JSON and decouple completely from display details to let the consumer handle it.

Structured output by default just makes sense for practically everything except `cat`. And while your markdown files or csv files might have quoted strings, looking at the raw files isn't something people really want from shells or editors.. they want something "rendered" in one way or another, for example with syntax highlighting.

Basically in 2025 neither humans nor machines benefit much from unstructured raw output. Almost any CLI that does this needs to be paired with a parser (like https://github.com/kellyjonbrazil/jc) and/or a renderer (like https://github.com/charmbracelet/glow). If no such pairing is available then it pushes many people to separately reinvent parsers badly. JSON's not perfect but (non-minified) it's human-readable enough to address the basic issues here without jumping all the way towards binary or (shudder) HTML

robot-wrangler··on The terminal of the future
> Acme:

Got a link to what you meant? This is pretty hard to search for.

> - Emacs

One thing in common with emacs, jupyter, vscode.. these are all capable platforms but not solutions, and if you want to replace your terminal emulator by building on top of them it's doable but doesn't feel very portable.

I'd challenge people that are making cool stuff to show it, and then ship it. Not a pile of config + a constellation of plugins at undeclared versions + a "simple" 12-step process that would-be adopters must copy/paste. That's platform customization, not something that feels like an application. Actually try bundling your cool hack as a docker container or a self-extracting executable of some kind so that it's low-effort reproducible.

robot-wrangler··on The terminal of the future
I think the modern unix'y thing to do here is always send all messages intended for the user to stderr (regardless of whether they are actually errors), and always respond with machine-friendly JSON on stdout by default. AFAIK there's no downsides to this.. you can still have color on stderr, you can still draw sixel pictures or whatever. Meanwhile pipes and redirection still work as expected.

No need for content/feature negotiations.. machine readable just defaults to JSON unless there's a --format flag for something else. And if you add that on the generation-side of the pipe, you just need to remember to put it on the consumer-side.

robot-wrangler··on Unexpected things that are people
> It does not much matter what outcomes that produces.

One outcome would be a predictable and mechanistic process, which reduces the potential for corruption and creates a more fair world. The currently popular legal theory in the US is far worse than "logical consistency" would be, because it's blatantly corrupt and autocratic. See Judge Barret's position on stare decisis (basically "should we honor precedent?") combined with reliance interests (basically "can we change anything without effecting someone?").

You know how division by zero allows you to prove 1=2? There's a similar thing at work when you allow completely contradictory legal systems to just continue with business as usual. Now a few people can do whatever they want with all the appearance of rigor/consistency/process without actually having any. As Leibnitz says, "let us calculate". Or just admit there is no process, and thus no real basis for the authority

robot-wrangler··on Unexpected things that are people
> It allows corporations to own property, enter contracts and appear as a plaintiff or defendant in lawsuits.

And yet these things do basically go all the way back to the Roman empire, and I'm sure the extent and privileges of corporate personhood have been litigated once or twice since then. If you disagree that

> corporate lawyers would like to pick from a bigger pool of personal rights in a perversion of the spirit of these original rights.

then what do you think they were working on?

robot-wrangler··on Unexpected things that are people
Personhood for non-persons is definitely absurd. But if you're actually stuck with a broken system, then the most logical thing to do is at least apply your broken logic consistently. That's an important part of the difference between rule of law and wild corrupt barbarism. Of course it's much better to actually fix absurdities, but if you can't or won't, inconsistency still has to be forbidden or else the whole thing is a farce
robot-wrangler··on Unexpected things that are people
Well I think one can justify it emotionally or logically. People identifying as anti-corporate are probably more likely to align as pro-environment. The emotional POV would be that non-person-personhood isn't good or bad intrinsically, it just depends if we approve of the area where the doctrine's applied.

The more logical reason is that if corporate personhood sucks and we have it anyway, then like it or not, now we need to extend it elsewhere just to level the playing field. If anti-environmental interests can hide behind it as a justification that makes their fight easier, then let the environmental interests do the same thing.

robot-wrangler··on Unexpected things that are people
Thought provoking. Most people who (quite reasonably) hate corporate personhood would probably have a knee-jerk reaction that personhood for a river can/should be normalized. The thing with temples stems ultimately from fairly practical matters if they hold such treasure, but it's a magnet for strife, and actually kind of surprising that in the case-study mentioned they resisted the opportunity to justify abuse of power. What is a lawyer really but a kind of priest or magician, changing material reality with obscure incantations of dubious origin?

Historically and practically speaking, I get the impression that the boat stuff seems the least controversial and makes the most sense. Incoherent to want to sue a river for flooding, but if a boat crashes into your house for example, then you'd like to be able to at least seize the boat without enduring the back-and-forth deflection between owners and operators.

robot-wrangler··on You should write an agent
I think it is misleading to suggest today that tool-calling for nontrivial stuff really works with local models. It just works in demos because those tools always accept one or two arguments, usually string literals or numbers. In the real world functions take more complex arguments, many arguments, or take a single argument that's an object with multiple attributes, etc. You can begin to work around this stuff by passing function signatures, typing details, and JSON-schemas to set expectations in context, but local models tend to fail at handling this kind of stuff long before you ever hit limits in the context window. There's a reason demos are always using 1 string literal like hostname, or 2 floats like lat/long. It's normal that passing a dictionary with a few strict requirements might need 300 retries instead of 3 to get a tool call that's syntactically correct and properly passed arguments. Actually `ping --help` for me shows like 20 options, and for any attempt to 1:1 map things with more args I think you'd start to see breakdown pretty quickly.

Zooming in on the details is fun but doesn't change the shape of what I was saying before. No need to muddy the water; very very simple stuff still requires very big local hardware or a SOTA model.

robot-wrangler··on You should write an agent
The expectation that reasonable people have isn't fully local claude code, that's a strawman. But it's also not ping tools or the simple weather agent that tutorials like to use. It's somewhere in between, isn't that obvious? If you're into evangelism, acknowledging this and actually taking a measured stance would help prevent light skeptics from turning into complete AI-deniers. If you mislead people about one thing, they will assume they are being misled about everything
robot-wrangler··on You should write an agent
> Not if you run them against local models, which are free to download and free to run .. run happily on CPU .. Cost isn't a reason not to explore this stuff.

Let's be realistic and not over-promise. Conversational slop and coding factorial will work. But the local experience for coding agents, tool-calling, and reasoning is still very bad until/unless you have a pretty expensive workstation. CPU and qwen 4b will be disappointing to even try experiments on. The only useful thing most people can realistically do locally is fuzzy search with simple RAG. Besides factorial, maybe some other stuff that's in the training set, like help with simple shell commands. (Great for people who are new to unix, but won't help the veteran dev who is trying to convince themselves AI is real or figuring out how to get it into their workflows)

Anyway, admitting that AI is still very much in a "pay to play" phase is actually ok. More measured stances, fewer reflexive detractors or boosters

robot-wrangler··on You should write an agent
I still don't really get it, but would like to hear more. Just to get it out of the way, there's obvious bad aspects. Re: press coverage, everything in AI is bound to be frustrating this way. The MCP ecosystem is currently still a lot of garbage. It feels like a very shitty app-store, lots of abandonware, things that are shipped without testing, the usual band-wagoning. For example instead of a single obvious RAG tool there's 200 different specific tools for ${language} docs

The core MCP tech though is not only directionally correct, but even the implementation seems to have made lots of good and forward-looking choices, even if those are still under-utilized. For example besides tools, it allows for sharing prompts/resources between agents. In time, I'm also expecting the idea of "many agents, one generic model in the background" is going to die off. For both costs and performance, agents will use special-purpose models but they still need a place and a way to collaborate. If some agents coordinate other agents, how do they talk? AFAIK without MCP the answer for this would be.. do all your work in the same framework and language, or to give all agents access to the same database or the same filesystem, reinventing ad-hoc protocols and comms for every system.

robot-wrangler··on You should write an agent
> Another thing to notice: we didn’t need MCP at all. That’s because MCP isn’t a fundamental enabling technology. The amount of coverage it gets is frustrating. It’s barely a technology at all. MCP is just a plugin interface for Claude Code and Cursor, a way of getting your own tools into code you don’t control. Write your own agent. Be a programmer. Deal in APIs, not plugins.

Hold up. These are all the right concerns but with the wrong conclusion.

You don't need MCP if you're making one agent, in one language, in one framework. But the open coding and research assistants that we really want will be composed of several. MCP is the only thing out there that's moving in a good direction in terms of enabling us to "just be programmers" and "use APIs", and maybe even test things in fairly isolated and reproducible contexts. Compare this to skills.md, which is actually defacto proprietary as of now, does not compose, has opaque run-times and dispatch, is pushing us towards certain models, languages and certain SDKs, etc.

MCP isn't a plugin interface for Claude, it's just JSON-RPC.

robot-wrangler··on Why aren't smart people happier?
Yes. A useful analogy is to imagine being an adult in a world populated only by children. Aside from the social alienation of it being hard to relate to others, there would be practical matters. The entertainment would all be tedious and predictable, all the rationalizations for bad behavior would be transparently self-interested. Enhanced capabilities for observation, prediction, and planning would make you a super-hero at problem-solving, but really, what does that get you except repetitive unfulfilling effort? Don't sweat the small stuff is good advice, but you couldn't actually ignore the futility. Don't focus on the negative is good advice, but in a world like that pessimism and realism are the same thing. Anyone would be miserable. The good-aligned person would likely withdraw or self-lobotomize. More cynical characters would harden their heart, seize power, and become king of all the blind babies and try to yoke them together and build a pyramid or something. (Yes, I've recently reread Understand by Ted Chiang, No a pyramid is not a plot point per se ;)

Thankfully the situation isn't actually this extreme, but I think what we're talking about is just a difference in degree and not a difference in kind. Seeing more clearly than others seems very uncomfortable at best, and frequently maladaptive and/or a recipe for derangement.

robot-wrangler··on What is a manifold?
Good game and a hard question, especially if you make "comparable" more explicit. I'd add "noncommercial, open-access", and "modern" in the sense that it happened under the current norms with respect to legacy and the social contract.
robot-wrangler··on What is a manifold?
I'm always surprised more people don't know about Quanta. Seems like it's currently the best science journalism out there, and IMO a very strong candidate for the single best place on the internet that's not crowd-sourced. The mixture of original art and technical diagrams is outstanding. Podcast is pretty good too, but I do wish they'd expand it to have someone with a good voice reading all the articles.

Besides not treating readers like idiots, they take themselves seriously, hire smart people, tell good stories but aren't afraid to stay technical, and simply skip all the clickbait garbage. Right now from the Scientific American front page: "Type 1 Diabetes science is having a moment". Or from Nature: "'Biotech Barbie' says ..". Granted I cherry-picked these offensive headlines pandering to facebook/twitter from many other options that might be legitimately interesting reads, but on Quanta there's also no paywalls, no cookie pop-ups, no thinly-veiled political rage-baiting either

robot-wrangler··on Visual Features Across Modalities: SVG and ASCII Art Cross-Modal Understanding
Generating and displaying diagrams in mermaid, svg, or css has become one of my go-to tests for reasoning. This seems fair because while SVG is admittedly syntactically difficult and maybe not emphasized in training, CSS is certainly a popular output target, and mermaid is very simple. It seems like SOTA should be able to draw and modify things that it "understands".

I'm much more interested in stuff like Venn diagrams and bipartite graphs than pictures of cats or pelicans riding bikes. It's similar to a code-generation problem in that output is a new artifact that's one step away from the problem-presentation, but it has the advantage that it's simpler than code, is less likely to have exact-match training data, usually has one correct answer, and is easy to check. Try making venn diagrams on a few circles with "exactly and only the following intersections" and gradually elaborating the spec.

This is a great way to get a starter diagram boilerplate if that's what you're looking for. One shot prompts for simple things are ok, sometimes. But it always completely falls apart when you try to iterate with small modifications, introducing errors in parts that were correct previously or ignoring requested changes. Maybe it's wrong to conclude anything from that, but to me this looks bad for the "they can reason!" argument and very bad for trusting complicated work in other domains that are harder to check. Haven't read TFA yet, but whether it confirms or denies my gut here hopefully it's going to add some perspective

robot-wrangler··on Why AC is cheap, but AC repair is a luxury
> everyone picks on radiology as the 'obvious' field that will get automated.

Nah, pre-LLM, I think the obvious fields to pick on were lawyers, middle-management, and "email jobs" generally. That was a big miscalculation, since most people (especially engineers) do not understand the politics of power. Those jobs tend to jealously protect the power that they have and systematically dismantle what accountability they might be subjected to. Engineers in general are much more likely to democratize (and thus threaten) power, by creating things like accountability via metrics, at the same time as they mostly refuse to unionize. Radiologists have some unions, medicine generally enjoys a moat of credentials and certification. Things that SWEs in particular rejected while they said "come on over, anyone can code". I doubt radiologists ever suggested themselves that they should be measured on throughput, but SWEs actually did push ideas of 10x engineers and metrics like lines-of-code for years to argue they are productive enough to deserve raises.

robot-wrangler··on Tell HN: X is opening any tweet link in a webview whether you press it or not
Obviously wrong. The typical user-hostile thing isn't this dumb, you'd see a teaser that's probably vaguely sexual and get some "sign up for the full experience" prodding. Literally any 2-person startup that's a week old would do better than this at being thirsty and awful
← PreviousPage 7 of 8Next →