HNHacker News
TopNewBestAskShowJobs

rsanheim

1,423 karma · joined March 31, 2008

https://rsanheim.com https://github.com/rsanheim
submissionscomments
rsanheim··on Agentic pelican on a bicycle
Did not feel LLM written to me - at least not overtly so. LLM editing/assisted perhaps?

It was a fun little post that felt accurate (ie confirmed my own biases ;)) about the current state of LLM models in a silly, but real, use-case.

The continual drive to out "llm written" articles feels a bit silly to me at this point. They are now part of the tools and tech we use, for better or worse. And to be clear, I think in a lot of cases it leans towards 'worse'.

But do you question if a video or photo was made with digital editing or filters or 'ai' tools (many of which we've had for years, just under different names) ? Do you worry about what tech was used in making your favorite album or song?

I get it, LLMs make it easy to produce trash content, but this is not a new problem. If you see trash, call it out as trash on its flaws, not on a presumption of how it was made.

rsanheim··on Guideline has been acquired by Gusto
because its a huge hassle that many financial services companies have no incentive to facilitate or make easy or discoverable. And for many folks a job change is a stressful event even in the _best_ of circumstances.

I know when I was laid off a week after covid lockdowns, the last thing I was thinking about was how to roll over my 401k as the market collapsed and I began interviewing and trying not to freak out.

having retirement and health benefits coupled to employment is antiquated and stupid, but changing tax code and finance system around 401ks is probably the least of our problems in the US.

rsanheim··on Fnox, a secret manager that pairs well with mise
Just...don't use them?

I've use mise happily for many months without using direnv or tasks, and everything I use it for works and is solid. Installs python, ruby, node, does the switching, does the shims, stays out of the way.

direnv and tasks and everything else mise can do is all opt-in.

rsanheim··on AWS multiple services outage in us-east-1
I wonder what kind of outage or incident or economic change will be required to cause a rejection of the big commercial clouds as the default deployment model.

The costs, performance overhead, and complexity of a modern AWS deployment are insane and so out of line with what most companies should be taking on. But hype + microservices + sunk cost, and here we are.

rsanheim··on I feel Apple has lost its alignment with me and other long-time customers
uhmm... wat?

where was there any mention in the OP of Jobs about diversity or his Syrian background?

rsanheim··on Claudia – Desktop companion for Claude code
Because Claude code does offer particular features. More important than “features” is the fact that it works and does the things you want like 60-70% of the time, with guardrails and practice and attention. Which is way better than competing tools.

Besides that. These tools are changing so fast that to build an agent agnostic tool would be insane given the speed and market pressures right now. Why support roo or cline or cursor cli if it adds 3x engineering cost for 20% more market reach? The reality is there are no standards around the way the actual leading tools work if you wanna build something on Claude/codex/(insert flavor of the week).

Gotta pick your horse and try to hang on, and hope you picked right.

rsanheim··on Emailing a one-time code is worse than passwords
The worst part about this is it just further reinforces horrible habits and expectations.

Using a modern password manager, like 1password, is _easier_, safer, and faster than the stupid email-token flow. it takes a little bit of work and attention at first to setup across a couple devices, and verify it works.... but its really about the same amount of effort as keeping track of a set of keys for your house, car, and maybe a workplace.

If you make a copy of a door key when you move into a new place, you test the key before assuming it works. Same thing with a password manager. Save a password on your phone, test it on a different device, and verify the magic sync works. Same as a key copier or some new locks a locksmith may install.

Humans can do this. You don't need to understand crypto or 2fa, but you can click 'create new password' and let the app save some insanely secure password for a new site. Same with a passkey, assuming you don't save to your builtin device storage that has some horrible, hidden user interface around backing that up for when your phone dies.

And the irony is the old flow just works better! You let the password manager do the autofill, and it takes a second or two, assuming their is an email _and_ a password input. Passkeys can be even faster.

rsanheim··on Jules, our asynchronous coding agent
Because their main business is selling ads and maintaining their stranglehold on that market via analytics, chrome, Chromebook, android, SSO via google, etc.

The dev focused products are a sideshow amongst different fiefdoms at google. They will never get first billing or focus.

rsanheim··on Tim Cook rallying Apple employees around AI efforts
Yeah, it turns out many nerds don't consider the fact that the amazing tools we are using to do constrained tasks aren't that great for more general purpose things. Writing a spike, spitting out unit tests, or vibe coding a front end feature is not the same as planning a trip to europe, balancing accounts, or managing a schedule.

So much attention, effort, and tooling has focused on getting llms better at writing more and more code. They can grep and curl and run scripts and iterate and build things really fast, and maybe even maintain it if given enough guardrails and direction.

But it turns out we have had a _ton_ of useful training data for models to work with for software. Not just books or docs, but examples, tests, snippets and full programs for just about any language. Show me a stackoverflow with playwright scripts or API calls (hah, as if thats possible) to build itineraries from delta, aa, united, priceline, expedia, etc, .... which is one part of one piece of the ai assistant pipe-dream.

I don't think its impossible as these tools get much smarter and more generally capable that we get decent assistants in other constrained, non-software domains, but it will take very good companies focusing on it for a long time. Much like any product that try to do these sorts of things.

Its so easy for programmers in our bubble to overlook the complexity involved in automating or even _describing_ simple tasks that humans navigate everyday via habit, learning, experience, and perception...all things that llms struggle with constantly.

rsanheim··on Tim Cook rallying Apple employees around AI efforts
The top folks at big ai labs and research can't even agree on what AGI means or if it is a useful term or taget. I guarantee the folks building opus 5 or GPT 6oMiniPlus5.5++ are not concerned about if the new model 'achieves AGI', regardless of how much marketing and PR folks want to throw that term around for every new release or blog post.
rsanheim··on Slack's 57MB 404 page
Absolutely insane. Some silly comparisons to what software has changed the world at less than 50MB:

- VisiCalc: ~27KB - i.e. the first killer app, and probably still the most successful. I wonder what DAUs on excel & google sheets compared to tiktok and insta are like...would be interesting to see.

- Wolfenstein 3D (1992): ~2MB - invented the FPS genre

- DOOM (1993): ~2.5MB (shareware), ~12MB (full)

- Quake (shareware) ~8.7 MB

- Quake (full) ~15-25 MB (excludes the CD audio)

- nginx: ~2-5MB - powers ~35% of all websites

- bash: ~1MB

- git (*nix only): ~3mb

- gcc + toolchain (ie ld, cc1, etc) ~40-45 mb

- FFmpeg: ~30-40MB

- curl: ~3MB

- OpenSSL: ~5MB

- windows 95: 50-55MB depending on features installed

- google.com circa 1998 - 10-15kb

- google.com today (anonymous session, ublock disabled) ~700kb

edit: my kingdom for markdown

rsanheim··on AI coding tools can reduce productivity
yes, definitely. https://github.com/BeehiveInnovations/zen-mcp-server is one example of people going off on this, but i'm sure there are many others
rsanheim··on AI coding tools can reduce productivity
I don't think that dichotomy is true at all, at least not with experienced software people.

Many folks I know are skeptical of the hype, or maybe full on anti/distrustful, due to reasons I think are valid. But many of those same people have tried llm tools, maybe chatgpt or copilot or cursor, and recognize the value even w/ huge misgivings. Some of have gone further with tools like claude code and seen the real potential there, quite a step beyond fancy auto-complete or just-in-time agents...but even there you can end up in rabbit-holes and drowning in horrible design.

In your incredibly reductive scale, I'm closer to 'love' than 'skeptical', but I'm often much of both sides. But I'd never write a prompt like 'write me some typescript' for any real work, or honestly anything close to that, unless its just for memes or demonstrations.

But no-one who programs for a living uses prompts like that, at least not for real work. That is just silly talk.

rsanheim··on Tools I love: mise(-en-place)
Mise is fantastic. Switched from asdf awhile ago and have not looked back.

I don't use the advanced task / env stuff, mostly just the tool management. Its been stable, fast, and gets out of the way.

rsanheim··on The Monster Inside ChatGPT
_try_ being the operative word here: https://www.npr.org/2024/04/02/1241148577/table-saw-injuries...

Sawstop has been mired in patent squatting and/or industry push back, depending on who you talk to of course.

rsanheim··on Why is the Rust compiler so slow?
Java still had slow startup and warmup time circa 2005-2007, on the order of 1-3 seconds for hello world and quite a bit more for real apps. That is horrendous for anything CLI based.

And you left out classloader/classpath/JAR dependency hell, which was horrid circa late 90s/early 2000s...and I'm guessing was still a struggle when Go really started development. Especially at Google's scale.

Don't get me wrong, Java has come a long way and is a fine language and the JVM is fantastic. But the java of 2025 is not the same as mid-to-late 2000s.

rsanheim··on A lumberjack created more than 200 sculptures in Wisconsin's Northwoods
I've lived in Wisconsin my whole life as well, minus 5 years in NC for a job.

The parent specifically mentioned _northwoods_ wisconsin, which is a different beast than Madison, Milwaukee, or some of the other cities.

So not sure how much of this hyperbolic reply to even take issue with, but it sounds like you've had a really rough go of it in WI. Sorry.

The rural / urban divide is real, though I think that is an America problem more than a WI problem.

Winter can suck, this is true. Though its changing a lot with climate change...and turns out the midwest may be one of the best places to be over the next 50 years w/ water shortages, fires, and hurricanes becoming more and more common everywhere else.

- Madison, Milwaukee, and surrounding areas both have fantastic restaurant and foodie scenes

- We have some of the best farmer's markets and local produce of anywhere, so if you like to cook you are in luck. There are great restaurants that will surprise you, even in smaller towns.

- I get gigabit fiber in Madison - I'm sure the situation is harder up north. I wouldn't expect getting fiber four hours north of SF would be easy, either. - The people here are generally friendly, hard-working, and solid...even if they vote or support politicians I find abhorrent. I don't think that is unique to rural WI.

- WI definitely has the drinking culture, but I find with most of my younger friends that habit is *much* less common. The drinking culture for kids now compared to the 1990s is massively improved and healthier.

- the music scene is alive and well, for both techno/house and underground punk/indie/whatever, which are the two scenes I've been in over the past 25+ years.

so yeah, wisconsin is a mixed bag (hell, so is cali and new york), but I think it still has a lot going for it.

I can tell you whenever I head home from yet another bay area trip, I may be dreading the WI winter but I'm definitely _not_ dreading the endless tech-babble you can't escape in any coffeeshop or restaurant around SF or LA. I'll take small talk around sports, fishing, or even the weather over crypto/VC/AI babble anyday.

rsanheim··on Generative AI coding tools and agents do not work for me
It's just Pattern matching. Most APIs, and hell, most code is not unique or special. Its all been done a thousands of times before. Thats why an LLM can be helpful on some tool you've written just for yourself and never released anywhere.

As to 'knows the answer', I'm don't even know what that means with these tools. All I know is if it is helpful or not.

rsanheim··on "The Illusion of Thinking" – Thoughts on This Important Paper
> Any physical process can be computed

Um, really? How so? Show me the computations that model gut biome. Or pain, or the immune system, or how plants communicate, or how the big bang happened (or didn’t).

And now define “physical” wrt things like pain, or sensation, or the body, or consciousness.

We know so little about so much! it’s amazing when people speak with such certainty about things like consciousness and computation.

I don’t disagree with your main point. I would just say we don’t know very much about our own consciousness and mindbody. And humans have been studying that at least a few thousand years.

Not everyone on HN or who works with or studies AI believes AGI is just the next level on some algorithm or scaling level we haven’t unlocked yet.

rsanheim··on Cursor 1.0
Same experience here. I got a Cursor Pro sub towards the end of 2024. Once Claude Code because available, my Cursor usage dropped _dramatically_ as I got up to speed on how Claude Code worked.

I still prefer Cursor for some things - namely UI updates or quick fixes and explanations. For everything else Claude Code is superior.

rsanheim··on My AI skeptic friends are all nuts
We will have the equivalent of Claude Sonnet 4 in a local LLM that can run well on a modern Mac w/ 36+ GB of ram in a year or two. Maybe faster. The local/open models are developing very fast in terms of quantization and how well they can run on consumer hardware.

Folks that are local LLMs everyday now will probably say you can basically emulate at least Sonnet 3.7 for coding if you have an real AI workstation. Which may be true, but the time and effort and cost involved is substantial.

rsanheim··on Run GitHub Actions locally
We've had open source CI. We still have it. I remember ye olde days of Hudson, before it was renamed Jenkins. Lotsa orgs still use Jenkins all over the place, it just doesn't get much press. It predates GHA, Circle, and many of the popular cloud offerings.

Turns out CI/CD is not an easy problem. I built a short-lived CI product before containers were really much of a thing ...you can guess how well that went.

Also, I'll take _any_ CI solution, closed or open, that tries to be the _opposite_ of the complexity borg that is k8s.

rsanheim··on High tariffs become 'real' with our first $36K bill
Not just small. _all_ businesses.
rsanheim··on OpenAI o3 and o4-mini
`ETOOMANYMODELS`

Is there a reputable, non-blogspam site that offers a 'cheat sheet' of sorts for what models to use, in particular for development? Not just openAI, but across the main cloud offerings and feasible local models?

I know there are the benchmarks, and directories like huggingface, and you can get a 'feel' for things by scanning threads here or other forums.

I'm thinking more of something that provides use-case tailored "top 3" choices by collecting and summarizing different data points. For example:

* agent & tool based dev (cloud) - [top 3 models] * agent & tool based dev (local) - m1, m2, m,3 * code review / high level analysis - ... * general tech questions - ... * technical writing (ADRs, needs assessments, etc) - ...

Part of the problem is how quickly the landscape changes everyday, and also just relying on benchmarks isn't enough: it ignores cost, and more importantly ignores actual user experience (which I realize is incredibly hard to aggregate & quantify).

rsanheim··on Local CI. Sign off on your own work
It isn't a matter of trust, its a matter of developers are humans. Humans make mistakes, and continuous integration helps catch mistakes.

I've been coding for 25+ years, and I sometimes I forget to remove debug output, or I forget to update an explicit exception I threw somewhere, or I forget to make sure dependencies build across all the platforms we need.

Running your build a second machine helps to catch a lot of those mistakes. It also helps enforce automation and good configuration practices and lint/formatting standards and all that, but fundamentally its helps us because humans make mistakes.

CI doesn't have to mean setting up a 12 step pipeline in GH actions with 3k lines of yaml and crazy triggers and workflows. It can just be an old dev machine (https://www.jamesshore.com/v2/blog/2006/continuous-integrati...) that pulls the latest code, runs `script/build`, and POSTs results somewhere.

Maybe local signoff works for 37signals, if so, thats great. But it isn't continuous integration.

rsanheim··on But what if I want a faster horse?
This.

Spotify hides the ability to just have your own stuff in favor of their algorithm they want to shove down your throat. It isn't just library management, but every part of their UI feels more like algo-driven exploration and recommendations as opposed to 'I'm in the drivers seat, and I decide when I want to explore'.

Its so frustrating, and I find it hard to explain to folks who have never had their own music collection finely curated - whether physical, in mp3s carefully renamed and organized (probably to be played thru winamp), or now in something like Roon or Apple music or one of the winamp-like local music apps.

Don't get me wrong, I've discovered great stuff via the various algo-radio things every service offers now and auto-plays if your queue ends. But I find much more great music via friends or music sites or reddit. And I want to be in the drivers seat when choosing music - starting with things I've added or favorited or added to playlists.

Anytime I have to use Spotify (typically to share w/ friends) its clear their product is all about _them_ and the stuff they want to push, and not about me.

rsanheim··on Game of Trees
I genuinely love this entry in the faq: https://gameoftrees.org/faq.html#pointless
rsanheim··on The Pain That Is GitHub Actions
CircleCI.
rsanheim··on The Pain That Is GitHub Actions
tldr but: don't use GitHub Actions. Its a mess, the availability is often atrocious, and the UI around it is _still_ as clunky as when they first rolled it out many years ago.

There are better solutions out there.

rsanheim··on IO Devices and Latency
That great infographic at the top illustrates one big reason why 'dev instances in the cloud' is a bad idea.
← PreviousPage 2 of 10Next →