Ask stories

Received the following email from Anthropic:

Hi,

Starting April 4 at 12pm PT / 8pm BST, you’ll no longer be able to use your Claude subscription limits for third-party harnesses including OpenClaw. You can still use them with your Claude account, but they will require extra usage, a pay-as-you-go option billed separately from your subscription.

Your subscription still covers all Claude products, including Claude Code and Claude Cowork. To keep using third-party harnesses with your Claude login, turn on extra usage for your account. This will be enforced April 4 starting with OpenClaw, but this policy applies to all third-party harnesses and will be rolled out to more shortly (read more).

To make the transition easier, we’re offering a one-time credit for extra usage equal to your monthly subscription price. Redeem your credit by April 17. We’re also introducing discounts when you pre-purchase bundles of extra usage (up to 30%).

We’ve been working to manage demand across the board, but these tools put an outsized strain on our systems. Capacity is a resource we manage carefully and we need to prioritize our customers using our core products. You will receive another email from us tomorrow where you’ll have the ability to refund your subscription if you prefer.

We built an end-to-end protein AI pipeline covering structure prediction, sequence design, and codon optimization. After comparing multiple transformer architectures for codon-level language modeling, CodonRoBERTa-large-v2 emerged as the clear winner with a perplexity of 4.10 and a Spearman CAI correlation of 0.40, significantly outperforming ModernBERT. We then scaled to 25 species, trained 4 production models in 55 GPU-hours, and built a species-conditioned system that no other open-source project offers. Complete results, architectural decisions, and runnable code below.

I am a beginner in game development.I am using Godot. For my game development ,I wanted to learn pixel art.I have downloaded Aesprite,watched a lot of videos,read a book on pixel art.Then ,I started to art but I can't draw properly.Will you please give me some guide/advice on learning pixel art?

What do you think?

Starting April 4 at 12pm PT / 8pm BST, you’ll no longer be able to use your Claude subscription limits for third-party harnesses including OpenClaw. Instead, they’ll require extra usage.

Your subscription usage still covers all Claude products, including Claude Code and Claude Cowork. To keep using third-party harnesses with your Claude plan, enable extra usage for your account. This will be enforced April 4 starting with OpenClaw, but this policy applies to all third-party harnesses and will be rolled out to more shortly (read more).

To make the transition easier, we’re offering a one-time credit for extra usage equal to $200 on your Team plan. Redeem your credit by April 17. We’re also introducing discounts when you pre-purchase bundles of extra usage (up to 30%) for your team.

We’ve been working to manage demand across the board, but these tools put an outsized strain on our systems. Capacity is a resource we manage carefully and we need to prioritize our customers using our core products. We appreciate your support as we make these changes.

Please state the location and include REMOTE for remote work, REMOTE (US) or similar if the country is restricted, and ONSITE when remote work is not an option.

Please only post if you personally are part of the hiring company—no recruiting firms or job boards. One post per company. If it isn't a household name, explain what your company does.

Please only post if you are actively filling a position and are committed to replying to applicants.

Commenters: please don't reply to job posts to complain about something. It's off topic here.

Readers: please only email if you are personally interested in the job.

Searchers: try https://nthesis.ai/public/hn-who-is-hiring, http://nchelluri.github.io/hnjobs/, https://hnjobs.emilburzo.com, or this (unofficial) Chrome extension: https://chromewebstore.google.com/detail/hn-hiring-pro/mpfal....

Don't miss this other fine thread: Who wants to be hired? https://news.ycombinator.com/item?id=47601858

Share your information if you are looking for work. Please use this format:

  Location:
  Remote:
  Willing to relocate:
  Technologies:
  Résumé/CV:
  Email:

Readers: please only email these addresses to discuss work opportunities.

There's a site for searching these posts at https://www.wantstobehired.com.

An email I just received from Claude:

We’re offering you a one-time credit for extra usage to your account, equal to your monthly subscription price. Redeem it here by April 17. Once claimed, it’s good for 90 days across Claude Code, Claude Cowork, chat, or third-party harnesses connected to your account.

You’re also now able to pre-purchase bundles of extra usage at up to 30% off. If you ever run past your subscription limits, this is the easiest way to keep going.

One note: starting April 4, third-party harnesses like OpenClaw connected to your Claude account will draw from extra usage instead of from your subscription. If you don’t use them, nothing changes. If you do, the credit and bundles above have you covered.

Thanks for building with Claude.

I’ve recently gone through several materials, including Antonio Gulli’s AI Agentic Design Patterns, Sam Bhagwat’s Principles of Building AI Agents and Patterns for Building AI Agents, as well as the courses from LangGraph Academy and some content on DataCamp.

This space is evolving very quickly, so I’m curious how others here are approaching learning. What resources, courses, papers, or hands-on approaches have you found most useful while building AI agents?

As user viewer (not channel owner) I have setup YouTube history to be only for max 3 months. Then auto-delete. Today I checked that this setting applies only to users' Video Watch History.

One has to manually go https://myactivity.google.com/page?page=youtube_comment_likes to delete Your Likes and Dislikes on YouTube and

delete comments at https://myactivity.google.com/page?hl=en&utm_medium=web&utm_source=youtube&page=youtube_comments for deleting Youtube Comments.

This is likely a bit unrealistic, but why can't we make a half rack server to go in someones basement that can also heat up their hot water and use the basement floor as a heat sink as well?

It seems like a lot of the blight of data centers is the energy to remove the heat. By distributing them into cool basements and even connecting them into the home heating system we could reduce that making them more efficient.

I was auditing a React site recently and noticed that the 'Wall of Love' widget added 400kb of JS and 3 iFrames, tanking the LCP. It felt wrong. We spend weeks optimizing every pixel, then we throw it all away for a slow, black-box testimonial slider.

I decided to build Reviewskits [0] to see if we could move to a purely headless, data-first approach using Bun and Hono.

The goal: Fetch raw JSON, build your own UI components, and keep a 100/100 Lighthouse score.

My question to HN: Do you prioritize the 'ease of use' of pre-made widgets over performance, or is the industry ready for a developer-first, headless infrastructure for social proof? I'm curious to hear how you handle this in your projects.

[0] https://github.com/reviews-kits-team/reviews-kits [1] https://reviewskits.com/

I know this sounds strange, and this could have been a reddit post, but I have realized ever since I went back to not rewriting things for polish using AI. I have had more positive feedback from the community and any personal/professional email.

people prefer reading short emails which doesn't have to be confusing.

people dont wanna read AI slop. It sounds nice sure, but lately no one cares about the english language being nice

everyone just wants to have a conversation/discussion with an actual human.

I personally am gonna just stop fixing grammar, and not care about it!

Thanks for listening to my rant. Written by a human, who's still in college, and might not get a job. :/ crying in the corner

Why? Obviously I have been accused of being a drunk, and it annoys me because I don't drink. I can drink if I feel others will consider it rude if I don't, but the last time I did that was, if memory serves, in 2014, and the last time before that probably somewhere before 2009. I don't much like to drink, and my parents were drunks so I consider it more insulting than one might otherwise.

A small background as to why people would think I was a drunk: I have a severely autistic and developmentally disabled son who we are pretty sure also has ADHD, and I am the only person who takes care of him (now I getting governmental help to watch him, but I still often have to spend days where I watch him from 7 in the morning to 10 at night and also if he wakes in the night overexcited and wants to run around the room yelling for a few hours), so I am very tired most of the time, also I have a few medical problems brought on from overwork in this issue that makes it sometimes painful when I walk.

So I understand why someone might think I was a drunk.

Background on why this has gotten me in a fighty mood: The people who accused me are his school, who know that I am his sole caregiver out of the government provided help. This was due to an incident that happened a few weeks ago in which, if there had been traffic coming, he would have been in trouble in the streets while under my care (I can describe more of this if you want but this is a pretty long ask)

They decided to make a report saying I picked him up while drunk. The report was full of half truths, and some potential lies, as well as the insult that I was drunk. Not only that but they claimed that I had been observed drinking beer at the local train station where there is a convenience store.

When I said you're mistaken I don't drink they stuck to their guns and would not apologize on that issue. When my ex-wife said "Out of all the things that Bryan has wrong with him, drinking is not one of them" they stuck to their guns.

Things I have done so far: I have gotten a notarized doctor's statement from my doctor since 2018 as to my medical ailments that might cause one to think this guy looks drunk right now, and that he has never seen any indication of a drinking problem in my blood samples and other labwork, in conjunction with that ever statement I have ever given for a medical questionnaire since he has known me says I do not drink.

I have statements from the convenience store workers that I have never bought any alcohol from them and that what I always buy are either hot coffees or triple shot espresso cans of Starbucks. One guy thought maybe the school thought it was beer because the cans are green, but they really aren't shaped like beer cans.

statements from work acquaintances and friends over the years that I have never shown an interest in drinking.

So what I would like to do now is as I asked, do an analysis of my credit card transactions from my bank records that I do not buy alcohol, doing so for the convenience story should be relatively easy, my transactions from them can be shown to match in amount the transactions they are willing to testify I do.

But I was thinking while it cannot be seen if I buy alcohol is there anything I can do to analyze my bank records to show that purchase history is not consistent with the expected purchase history of someone with a drinking problem? Obviously I do not go to bars.

Unfortunately I do not have credit card transactions saved from shopping so I can't show I have never bought alcohol, but I can start saving this month, and that would show I do not buy currently and could be taken to show that there are no anomalies with current months.

Obviously this might seem like an overreaction, but I dislike getting accused of drunkenness, lied about, and being surveilled incompetently enough that the surveillance assumes a can of Starbucks is beer.

What's a simple, lightweight IDE that I can use to inspect my code while CC runs? I don't need much functionality at all - but I do need Remote Access since most of my work is in a remote cluster.

The IDE just lets me 'read only' my code and double check things, while all my changes are through Claude's interface.

On a newest version, I attempted to download newest yt-dlp only to be warned of "Suspicious Download". No explanation what that means was provided.

At the beginning it was very cool to see them write a lot of code for me, in particular while I was ramping up on a new codebase. But I am seeing myself using agents less and less as I get more acquainted with the codebase. I might use them as a glorified copy n paste or to get a tool running. And they're indeed very useful to get the gist of a part of the code I don't fully grasp. But I am back to writing I'd say 90% of the code myself. Is this your experience as well?

Postgres has relied on the OS to handle text sorting for most of its history. When glibc 2.28 shipped in 2018 with a major Unicode collation overhaul, every existing text index built under the old rules became invalid... but silently. No warnings, no errors. Just wrong query results and missed rows.

Postgres 17 added a builtin locale provider that removes the external dependency entirely:

initdb --locale-provider=builtin --locale=C.UTF-8

This change helps sorting to become stable across OS upgrades. glibc is still the default in Postgres 18, so this must be specified when creating a new cluster.

For clusters already running: Postgres 13+ will log a warning when a collation version changes. That warning is an instruction to rebuild affected indexes.

Get more details here in this week's PG Phriday blog post from Shaun Thomas: https://www.pgedge.com/blog/what-is-a-collation-and-why-is-my-data-corrupt

Curious to find those hidden gems that boost productivity or make dev work more efficient. Mired in my own processes and need a change/shift. I'm hoping there's still some non-AI stuff out there that's delightful to use (in a nerdy sense).

Many of the prescription drugs as far as I know come from China and India. Should trade be cut off from China and India:

- how long would each country have until they run out of prescription drugs? pharmacy, lab and national guard stockpiles

- how long would it take each country to ramp up their own production of said drugs?

- would some countries just let their supplies run out? Eugenics, allowing blood pressure rebounds and strokes among other things to run their course to reduce long term health care costs.

just something that crossed my mind after listening to Doomberg for a while.

Curious what HackerNews users are using right now. Mapping my IDE usage since 2022

Goland (2022-2024)-> Cursor(November 2024 to February 2026) -> Claude Code (& VSCode or Cursor for manual edits)

The Claude Code setup is interesting, I use the terminal or GitHub for diffs. I do open an editor to do manual edits, especially when I am doing something new( that the LLM hasn't been trained on) or debugging something.

Potential improvements

Stripe Projects ( creating API keys from CLI) as something that I have wanted in the past as with LLMs sometimes with a project the slowest part seems to be deployment / bringing all the keys. I also don't enjoy the fact that I can't push a job to the web when I am leaving work. Worktrees aren't fun as I can't run the services on different ports that easily for testing, further managing different TMUXs is painful. I am curious what tools people here are using, also what do people do while the LLM generates.

I have built a pipeline that takes construction documents as PDFs, extracts out the floor plans.

It extracts out BIM data from the plans at pixel accuracy producing data on walls, floors, windows and doors accurately.

It also generates 3D impressions and a brief 10s 3D video of the floor plans.

I have had multiple discussion to build within a company already solving the same problem but not closed a deal yet and considering taking it to the market as a solo founder.

I am outside the UK and USA and was seeking a founder for market access and future fundraising.

What is the best advice on how to go to market as a sole founder?

I can't believe this After a very tedious process to get my business account fully verified, I was finally happy to be able to use stripe services but alas..

I invoiced a client for some software service, about 6 or 7 invoices, after they paid them, stripe decided that my account was high risk and suspended it, even after asking for a review, they didn't change their mind and said the transactions will be refunded

They refunded all of the invoices but one worth $3.5k or so. I tried to contact their support but got no response, so I sent a complaint through their complains form, they responded with the following: "Summary: You have contacted Stripe to express your concern regarding the closure of your account.

Details of our review: We reviewed your account further, and we’ve determined that we still won’t be able to support [company name] moving forward. The closure of your account is in accordance with Stripe's Services Agreement: https://stripe.com/ae/legal/ssa#termination

Note that, in general, refunds on eligible card payments are issued within 5 days of the date of the account rejection. After the 5-day period, we will not be refunding additional charges. If a balance still remains in your account after all reversals have been processed, it will not be made available to you, in accordance with the Service Terms for Stripe Payments.

We know that this is disappointing news, and we're sorry that there's nothing more we can do."

--

What can I do? My business is registered in the UAE

Disclaimer: I run a low code service and still many customers use us. But am curious what’s the developer /cto perspective on this?

We do have a AI assistant to help build on top of platform but the primary appeal of low code was don’t bother about code and care about the app which with agentic sw development may also be achieved.

Paying $200 a month, I hit my weekly in 3 days last week.

Today, first day of the new week and I've hit 20% of the weekly limit in about 2 hours. At that rate, I'll hit the weekly limit in 10 hours of work. I haven't changed anything about the way I'm working since previous months and I've never hit limits like this before.

Clearly, we're being gaslighted and it's time to move on.

For an equivalent budget, what's the next best balance of quality and session limits for agent assisted coding?

I found that if your post is a LLM generated content, it will be auto-hide.

And I heard long time ago for same product you could not share in Show HN for multiple times.

What is the mechanism behind? Is Hacker News using LLM to judge each post or retrieve the product name and deduplicate in Show HN?

I've used https://european-alternatives.eu in the past but I've found it impossible to submit anything there (I've got one submission "Waiting for Review" since July 2025), and I haven't really seen updates in categories/apps for at least 6 months, which is, at a minimum, strange.

I've also reached out to the owner a couple of times (via chat and via email) over the last year and never got a reply.

Has anyone else had a similar experience? If so, is there any other kind of listing you look for EU alternatives to US software/hardware?

Thanks!

EDIT: My motivations are my own, this isn't supposed to be a political or social commentary of any kind. I happen to prefer to spend money in "local companies" whenever possible, and I'm in the EU.

Would you like to participate in a study on AI’s impact on software development? We are researchers at New York University and City, University of London conducting an interview study on how new AI tools are changing the work of software developers. We are looking to speak with developers of all seniority levels, including those in leadership roles, who can share their experiences and perspectives on using (or choosing not to use) AI in their day-to-day work.

Interviews will last 45 to 60 minutes and take place via Zoom. Participants will be asked about their workflow, AI tool usage, and how their role has evolved over time. All responses will be kept confidential and used for academic research purposes only. Research participants need to be based in the U.S.

If interested, please fill out this brief form so that we can contact you: https://nyu.qualtrics.com/jfe/form/SV_cHkvoczxgtaLLo2

Thank you!

While running llama.cpp on a Samsung Galaxy Watch 4 Classic (armeabi-v7a, Mali G68), I noticed the Vulkan backend was rejecting every quantized MUL_MAT operation despite reporting "33/33 layers offloaded to GPU".

Root cause: a missing block size division in tensor stride calculation inside create_tensor() in llama-model-loader.cpp. The wrong stride cascades into ggml_nbytes() overflow, exceeding max_buffer_size on 32-bit where size_t is 32-bit.

On 64-bit devices the overflow is silently masked — wrong value but still within GPU memory limits so nobody noticed. Bug has likely been there for years.

Fix and context: https://github.com/Perinban/llama.cpp/tree/axon-dev

I've been building a third-party cross-platform Discord client called kind (kind is not discord) in C++ and Qt 6. No Electron, no web wrapper. A proper native app for Windows, macOS, and Linux.

The UI always loads instantly. On first load, things take as long to appear as Discord's network responses. On subsequent loads, everything comes from cache and is validated against REST in the background. It feels as fast as opening a plaintext file.

It's further along than I expected when I started. What's working right now: gateway with exponential backoff and reconnect, REST with per-route rate limiting, SQLite-backed cache with async reads/writes, full Discord permission resolution, and a composable block renderer that handles every Discord message type including embeds, attachments, reactions, stickers, and components. There's a full markdown parser, an async image cache with memory LRU and disk persistence, an unread and mute state system that survives restarts, and 370 passing unit and integration tests.

Voice is a post-launch problem. So are Nitro features.

Is there actually an audience for this outside of Linux power users? I know that's where the pain is most acute, but I'm genuinely curious whether Windows users would switch for the performance alone or whether the official client is good enough there that it doesn't matter.