nlh

I love that you’re tackling this problem, and congrats on launching and getting this on HN!

This does feel like a real problem. The thing that concerns me (and likely other devs here) is that it adds an additional remote API dependency for a very core part of a system when a lot of people are trying to keep those dependencies to an absolute minimum. When your service goes down (not if), everyone who’s dependent on you will not be able to register new users, etc.

Is there any way you can offer this as a library instead? You deserve to get paid of course - maybe provide the library and initial data and charge for updates / premium checks, something like that.

That's an unhelpful and unnecessarily nasty comment. Millions and millions and millions of people trust Apple. Whether you agree or not, to say they "deserve" something like this for doing what any normal person on this earth would do (and is marketed-at to do) is obnoxious.

“In one court case in Ohio, Dollar General’s lawyers argued that “it is virtually impossible for a retailer to match shelf pricing and scanned pricing 100% of the time for all items. Perfection in this regard is neither plausible nor expected under the law.””

Sorry—-what? Isn’t that one of the fundamental basic jobs to be done and expectations of a retailer? You put physical things on display for sale, you mark prices on them, and you sell them. When the prices change, you send one of your employees to the appropriate shelves and you change the tag.

When on earth did we get into a world where that absolutely fundamental most basic task is now too burdensome to do with accuracy?

I’ve never understood this - it’s maddening. I grew up in the US and the bare minimum was always at least a shower curtain (inner and outer), and if not that, a proper door.

Why on earth did this half-pane of glass become standard in so many places. It’s completely ineffective and ends up with water everywhere.

What's remarkable to me is how deep OpenAI is going on "ChatGPT as communication partner / chatbot", as opposed to Anthropic's approach of "Claude as the best coding tool / professional AI for spreadsheets, etc.".

I know this is marketing at play and OpenAI has plenty of resources developed to advancing their frontier models, but it's starting to really come into view that OpenAI wants to replace Google and be the default app / page for everyone on earth to talk to.

Man, maybe I'm getting old and jaded, but it's not often that I read a post that literally makes my skin crawl.

This is so transparently icky. "Oh woe is us! We're being sued and we're looking out for YOU the user, who is definitely not the product. We are just a 'lil 'ol (near) trillion-dollar business trying to protect you!"

Come ON.

Look I don't actually know who's in the right in the OAI vs. NYT dispute, and frankly I personally lean more toward the side the says that you are allowed to train models on the world's information as long as you consume it legally and don't violate copyright.

But this transparent attempt to get user sympathy under insanely disingenuous pretenses is just absurd.

You ask a great question. My sense is that the engineers fell into three camps (as they do here on HN as well):

1) I don’t really like these AI tools I write better code anyway and they just slow me down

2) I like these tools they make me 10% faster but they’re more like spell check / autocomplete for me than life-changing and I don’t want to go all in on agentic coding, etc and I still want to hand write everything, and:

3) I am no longer writing code, I am using AI tools (often in parallel) to write code and I am acting like an engineering manager / PM instead of an IC.

For better or for worse, and there is much to debate about this, I think he wanted just the (3) folks and a handful of (2) folks to try and salvage things otherwise it wasn’t worth the burn :(

Haven't RTFA (paywall) but an anecdote:

I know a startup founder whose company is going through a bit of a struggle - they hired too many engineers, they haven't gotten product-market fit yet, and they are down to <1 year of runway.

The founder needed to do a layoff (which sucks in every dimension) and made the decision to go all-in on AI-assisted coding. He basically said "if you're not willing to go along, we're going to have to let you go." Many engineers refused and left, and the ones that stayed are committed to giving it a shot with Claude, Codex, etc.

Their runway is now doubled (2 years), they've got a smaller team, and they're going to see if they can throw enough experiments at the wall over the next 18 months to find product-market fit.

If they fail, it's going to be another "bad CEO thought AI could fix his company's problems" story.

But if they succeed....

(Curious what you all would have done in this situation btw...!)

Oh man this definitely makes me wax nostalgic for that golden era ... it was 2013-2016 for me. I would throw an annual holiday party w/ my roommate in SF every year and I recall being able to just go down the list of my FB friends and click "invite, invite, invite" and everyone I cared about would show up and we all had a wonderful time. Sigh.

The only thing missing here is a way to SEND emails from your new business email. I don’t think CloudFlare (yet) offers virtual SMTP for your routed email, so you can receive emails to your new “noah@mynewcompany.com” address but sent mail will still come from your personal email address.

Anyone have a simple way of solving this? I’ve done it in a complex way (setting up my own SMTP service through through Google Apps, a paid service), but I’m wondering if there’s a simpler approach I should use going forward for these quick setups.

It's funny and this timing is perfect - just today two (mostly non-technical) friends, unprompted, sent me two different "projects" they just launched. Both have the same identical and unmistakable AI design with pills and tags and highlights and icons and all that. Both are bland, don't really offer anything deeply novel, but are unmistakably real and perhaps even usable.

On the one hand - I'm actually psyched that mostly nontechnical people have successfully launched their ideas into the world. This would have eaten up time and at least several $thousand previously.

On the other hand, they are both super blah projects.

Date I say it: The AI Slop Web App era has arrived in full force.

Oh for sure - APMEX is selling plenty of gold and they're doing great. But they buy gold too and ship it off to refiners just like any dealer. And yes there is plenty of demand for gold - but my only point is that demand right now is for Gold The Commodity (eg a hedge fund / government / etc.) who just want the financial instrument (that happens to be backed by physical bricks in a warehouse...they hope lol) as opposed to the retail buyer who wants gold the way they want Crypto or Pokemon cards.

For example, one dealer I know last month sold $5m in physical gold coins (to consumers who wants gold) but they bought $30m in physical gold coins to be shipped off and melted. They, like APMEX, make money on both sides so they're plenty happy. But a 6:1 buy:sell ratio is pretty wild.

That’s certainly part of it (and the argument most gold bugs make as to why we should all be buying gold), but gold has gone up 133% in the last 3 years which is wayyyyy beyond inflation tracking.

Background: I’m a full-time rare coin dealer (and software engineer, which is why I’m here haha).

One of the fascinating things happening in this market is that there’s a massive disconnect between the spot price of gold ($4000), which is driven and set by massive trades and hedges on international commodities exchanges, and the supply and demand for actual physical gold - the kind that’s sold at Costco in bars or by the US mint in 1oz coins.

Since so much activity is driven by futures contracts online (without physical settlement), this disconnect means that the average consumer more or less is either holding physical gold they already have or selling physical gold they already have to cash out at these historic levels. There just aren’t a lot of regular people out there who are saying “yeah I want lots of gold at $4000/oz”.

So brick and mortar dealers are being flooded with gold in their shops these days, and the supply chain looks pretty wild - tons of that physical gold is going to refineries who are melting it down and shipping it off as physical collateral for the futures contracts - but the big buyers out there don’t actually want or need all this expensive physical gold. The price spike isn’t because all of a sudden there’s increased demand for physical gold - it’s just a financial instrument at this point that happens to have a physical counterpart.

None of this is to say there’s a problem with this - it’s just fascinating that net result of this huge spike is lots of jewelry and old coins being melted into bricks of gold that will sit quietly in secure warehouses and vaults around the world.

One fantastic tip I discovered (sorry I've forgotten who wrote it but probably a fellow HNer):

If you're using an AI for the "architecture" / spec phase, play a few of the models off each other.

I will start with a conversation in Cursor (with appropriate context) and ask Gemini 2.5 Pro to ask clarifying questions and then propose a solution, and once I've got something, switch the model to O3 (or your other preferred thinking model of choice - GPT-5 now?). Add the line "please review the previous conversation and critique the design, ask clarifying questions, and proposal alternatives if you think this is the wrong direction."

Do that a few times back and forth and with your own brain input, you should have a pretty robust conversation log and outline of a good solution.

Export that whole conversation into an .md doc, and use THAT in context with Claude Code to actually dive in and start writing code.

You'll still need to review everything and there will still be errors and bad decisions, but overall this has worked surprisingly well and efficiently for me so far.

What I have found Claude Code is extremely good at is that it makes one change at a time, gives you a chance to read the code its changing, and lets you give feedback in real time and steer it properly. I find the mental load with this method to be MUCH lower than Cursor or any of the other tools which give you two very different options: "Ask" mode which dumps a ton of suggestions on your and then requires semi-manual implementation, or "Agent" mode which dumps a ton of actual changes on you and requires your inspection and feedback and roll-backs, etc.

This may not work for everyone, but as a solo dev who wants to keep a real mental model of my work (and not let it get polluted with AI slop), the Claude Code approach just works really well for me. It's like having a coding partner who can iterate and change direction as you talk, not a junior dev who dumps a pile of code on your plate without discussion.

Can anyone recommend a workflow / tools that accomplishes a slightly more augmented version of antirez’ workflow & suggestions minus the copy-pasting?

I am on board to agree that pure LLM + pure original full code as context is the best path at the moment, but I’d love to be able to use some shortcuts like quickly applying changes, checkpoints, etc.

My persistent (and not unfounded?) worry is that all the major tools & plugins (Cursor, Cline/Roo) all play games with their own sub-prompts and context “efficiency”.

What’s the purest solution?

Awesome! I went KB2NDR -> AA2KT (because at the time AA was the hot new prefix for Advanced) and then when vanities came out I grabbed a 1x2 that I’d dreamed of getting had I been licensed in the 70s :)

If I may wax nostalgic for a moment:

I got my original Novice license in 1990 I think (age 12). I was SO excited to participate in my local club’s Field Day outing - but it was a nail biter for my physical license to arrive in the mail in time.

I’d somehow gotten the word from the FCC that my license was issued and I got my call sign (KB2NDR) but I was so worried it wouldn’t get to my house in time for the weekend (and it didn’t!) - but the club president was super chill and said “I trust ya” so he let me participate and I still remember every minute of that weekend to this day.

It was my first full-fledged ham event (my first all-nighter too), sitting in the tents working on HF rigs I could only dream of affording at 3am with guys chain smoking cigars. They were probably chuffed at this nerdy 12 y/o who wanted to play along (minus the smoking bit).

“…CQ Field Day CQ Field Day this is K2-Zed-O, K2-Zulu-Oscar…”

(That weekend launched my short intensity but long lived ham life. Leveled up to Extra and I still have K2KD active today but haven’t touched a radio in years)

^^ this. Also, not only burn the session - sometimes you need to burn Claude. I'll often find the same problem that Claude struggles with is pretty easy for O3 + GPT 4.1 (or vice versa).

I'm trying to think about why this bothers me so much. I'm a happy (I think?) user of Warp The Terminal (eg their product with all of the AI features turned off), and it's a really nice terminal.

I think it bugs me because it feels like a company with a product in search of a problem, trying to maximize returns for their investors, cashing in the AI hype, instead of just trying to build a Really Good Product.

Their terminal is Really Good! And in fact an AI-enabled terminal could even be Really Good - but they won't of course let you use your own API keys (because $markup$), and now they're pitching this whole "ADE" thing which just screams to me "Hey Cursor got a huge valuation so what's our secret sauce?".

And look, I can't fault them -- they're doing exactly what VC-backed companies are built to do. I would probably be doing precisely the same thing if I were CEO.

But it still bugs me.

I'll never forget the run-in I had with this corner of the labor market:

Back about 15 years ago I was running a small business in the auto industry. A guy who did deliveries for us (entire job: driving cars to people) who was in his early 20s found himself in a very minor fender-bender -- he rear-ended someone else. He claimed his twisted his ankle from the jolt (based on where he was resting his foot). Fine, no big deal.

He went to see a doctor shortly afterwards and immediately filed for a Worker's Comp claim. He then kept seeing that doctor and within a few weeks was given...permanent disability. Literally got a doctor to say he'd never be able to work again. Full sign-off. He of course was seen walking around just fine a few months later.

Absolute insanity.

Could be an issue with your mic (clogged? weak connection?). I just installed and used it and the app works perfectly for me and picks up birds that are calling faintly and in the distance.

Like a caveman! :)

I actually haven't explored the inline autocomplete space that much yet. I really like Cursor Tab (or Supermaven or even plain 'ol Copilot) -- they all seem really good. I'm sure there's a Cline equivalent out there, but I don't know what it is (yet).

I've played with Shopify for many years and it just can't do some of the low-level stuff I want it to do without badly mangling how product taxonomy works. Coins are a different kind of beast than typical retail and if I'm going to be writing code anyway, I'd rather write my own code (vs. writing code "on the shopify platform").

The really sensitive stuff - financial stuff - they handle through checkout (and I don't need anything special there), so I've really just written my own catalog & front-end.

Cursor changed their pricing recently and now charge a 20% markup on LLM API calls to use their "Max" models (which from what I gather are the full extent of the LLM context windows you get in the API anyway).

I also don't love that Cursor generally plays "context compression" games since they have an incentive to keep their costs minimal. I just don't love any of these tools that try to be a bit too smart as a middleman between you and the LLM (where smart is often defined as 'try to save us the most money or be maximally efficient without the user noticing').

Cline also tries to be smart, but it's all for the benefit of the user. I like the transparent pricing -- you bring your own API key so you're paying the underlying API costs without a middle man markup.

Am I being pennywise and should I just use Cursor directly? Maybe...but I've found my Cline results to be generally better for the more complex queries.

Ok this is great and perfect timing -- I've been playing around with Warp (the terminal) and while I love the idea of their terminal-based "agent" (eg tool loop), I don't love the whole Cursor-esque model of "trust us we'll make a good prompt and LLM calls for you" (and charge you for it), so I was hoping for a simple CLI-based terminal agent just to solve for my lack of shell-fu.

I am keenly aware this is a major footgun here, but it seems that a terminal tool + llm would be a perfect lightweight solution.

Is there a way to have llm get permission for each tool call the way other "agents" do this? ("llm would like to call `rm -rf ./*` press Y to confirm...")

Would be a decent way to prevent letting an llm run wild on my terminal and still provide some measure of protection.

I use Cline within Cursor — best of both worlds!

I'm a tech guy turned rare coin & currency dealer -- this is my world:

https://rarity7.com

The retail site is 100% custom code built in Crystal (server) and Svelte (client). The only part that isn't running my own code is our checkout flow -- I let Shopify handle everything after "Add to Cart".

Our system backend is a separate Crystal app which handles inventory management, pricing research, and price prediction. I've developed an ML model to do price prediction and it kinda works?

What I'm actually working on: This is my full-time gig and probably 60% of my time is spent running the business (going to coin shows, buying coins, photographing new purchases, etc.) and 40% is spent writing code to make the 60% run more efficiently :). It seems I have an infinite list of things to do -- improvements to our retail site; Improvements in how to efficiently go from coin to retail listing (turns out you can send just photos of coins to Claude and with the right prompt it will actually give you a reasonably good description that doesn't sound toooooo AI slop-y); Next "big" project is adapting our ML model for paper currency. The taxonomy is similar but not the same and there's a whole world of notes out there that need to be priced.

Always happy to talk about this stuff so always feel free to email with any numismatic (or tech-numismatic) questions. noah@rarity7.com.

Love this. As a new(ish) dad to a 16-month-old little girl, we're not in the YouTube vortex yet, but I know it's inevitable. When it comes time, I want to balance "she can watch and learn stuff" against my general sentiment against screen time / devices (which we've been pretty good at so far).

Anyway, a long way of saying awesome - would love to be on your list. I'll send you an email separately.