chw9e

This happened to me several years ago. I got signed up to probably 700 newsletters overnight. In the middle of all of the sign ups there was activity on my airbnb account where my notification settings were changed. when i checked my airbnb i noticed that someone had created a fake listing under my account and disabled booking notifications for it. a real multi-layer scam where the hacker would be making money off a fake listing on someone else's account who would probably never even realize it.

As someone who never used Phind, how was it better than copilot? Did it integrate data from broader sources or just do a better job of presenting it? Was it faster?

Oh, I just picked Anthropic’s Sonnet4 to start because it’s really good at agentic stuff and because of the subscription benefits.

I will add more model support soon - any models you particularly want to see?

I’m also hoping to test out RL on the tools to get a fine-tuned model specifically for browser automation eventually.

It also has its own cli if you just want to run it alone too

Yea it can be, but you will pay Anthropic API price. If you already use Claude code on a subscription then you can reuse your subscription for this MCP, making it effectively free

this is t3 chat from what i understand, but probably many people already doing this. this is a good approach for wrappers.

probably the idea behind the coding tools eventually. cursor charges a 20% margin on every token for their max models but people still use them

Think we've got a long time yet for that. We're going to be writing code a lot faster but getting these things to 90-95% on such a wide variety of tasks is going to be a monumental effort, the first 60-70% on anything is always much easier than the last 5-10%.

Also there's a matter of taste, as commented above, the best way to use these is going to be running multiple runs at once (that's going to be super expensive right now so we'll need inference improvements on today's SOTA models to make this something we can reasonably do on every task). Then somebody needs to pick which run made the best code, and even then you're going to want code review probably from a human if it's written by machine.

Trusting the machine and just vibe coding stuff is fine for small projects or maybe even smaller features, but for a codebase that's going to be around for a while I expect we're going to want a lot of human involvement in the architecture. AI can help us explore different paths faster, but humans need to be driving it still for quite some time - whether that's by encoding their taste into other models or by manually reviewing stuff, either way it's going to take maintenance work.

In the near-term, I expect engineering teams to start looking for how to leverage background agents more. New engineering flows need to be built around these and I am bearish on the current status quo of just outsource everything to the beefiest models and hope they can one-shot it. Reviewing a bunch of AI code is also terrible and we have to find a better way of doing that.

I expect since we're going to be stuck on figuring out background agents for a while that teams will start to get in the weeds and view these agents as critical infra that needs to be designed and maintained in-house. For most companies, foundation labs will just be an API call, not hosting the agents themselves. There's a lot that can be done with agents that hasn't been explored much at all yet, we're still super early here and that's going to be where a lot of new engineering infra work comes from in the next 3-5 years.

Yea, Airbnb keeps trying this services idea but not sure it's the right fit.

I haven’t yet come across an agent that can write beautiful code.

o3 in codex is pretty close sometimes. I prefer to use it for planning/review but it far exceeds my expectations (and sometimes my own abilities) quite regularly.

The things I'm usually tabbing through in cursor are not the things that make me feel a lot of enjoyment in your work. The things that are most enjoyable are usually the system level design aspects, the refactorings to make things work better. These you can brainstorm with AI, but cannot delegate to AI today.

The rest is glorified boilerplate that I find usually saps me of my energy, not gives me energy. I'm a fan of anything that can help me skip over that and get to the more enjoyable work.

How good were LLMs at building code that uses LLMs? This used to be a pain in previous versions because the LLM didn't have any LLM-using code in its training data. And what was the rough breakdown of LLM usage in coding? (looking for something like 20% claude code, 30% gemini 2.5, the rest manual)

This was a very cool paper about using diffusion language models and beam search: https://arxiv.org/html/2405.20519v1

Just looking at all of the amazing tools and workflows that people have made with ComfyUI and stuff makes me wonder what we could do with diffusion LMs. It seems diffusion models are much more easily hackable than LLMs.

Hey HN - I made a bunch of GenAI projects in 2024:

- 2 frontend React/Tailwind codegen projects (1 an agent, and 1 a really cool website builder)

- 1 Node.js codegen AI agent

- 1 ecommerce product placement image generator

I'm not working on these anymore (and the code and dependencies are probably far out of date now unfortunately - haven't looked at some of these for a year at this point).

But I thought I would just go ahead and make the code public and share it out there in case it can help anyone or inspire some new ideas.

Besides just sharing the code, if you click through to the detailed pages for each project you can see a demo video of each one showing how it works and read some of my notes about each one.

About the projects:

I really love the website builder, it's probably my favorite project I've done - so many cool details.

Other projects have some cool agent things going on, maybe some novel approaches to codegen in there - not sure.

The image generation one has a bunch of ComfyUI workflows that I spent a bunch of time on.

Happy to answer any questions here - although I'll probably have to review the code if anything is too in the weeds as I've forgotten a lot already.

Also the code is not very well-organized and there's some dead code and stuff floating around and things commented out or half-implemented at places. I didn't bother to clean it up too much besides making sure I'm not leaking any env secrets or API keys (if you spot one, let me know please!)

And last - these are mostly written in Node.js, React, and a bit of Python. I don't claim to be an expert in these languages and was pretty unfamiliar with using them before I built these - so don't use this as a reference on how to write great code, rather I hope you enjoy the concepts behind the features.

Hope it can be of interest and potentially of help to someone out there!

Hey HN - I made a bunch of GenAI projects in 2024: - 2 frontend React/Tailwind codegen projects (1 an agent, and 1 a really cool website builder) - 1 Node.js codegen AI agent - 1 ecommerce product placement image generator

I'm not working on these anymore (and the code and dependencies are probably far out of date now unfortunately - haven't looked at some of these for a year at this point).

But I thought I would just go ahead and make the code public and share it out there in case it can help anyone or inspire some new ideas.

Besides just sharing the code, if you click through to the detailed pages for each project you can see a demo video of each one showing how it works and read some of my notes about each one.

About the projects:

I really love the website builder, it's probably my favorite project I've done - so many cool details.

Other projects have some cool agent things going on, maybe some novel approaches to codegen in there - not sure.

The image generation one has a bunch of ComfyUI workflows that I spent a bunch of time on.

Happy to answer any questions here - although I'll probably have to review the code if anything is too in the weeds as I've forgotten a lot already.

Also the code is not very well-organized and there's some dead code and stuff floating around and things commented out or half-implemented at places. I didn't bother to clean it up too much besides making sure I'm not leaking any env secrets or API keys (if you spot one, let me know please!)

And last - these are mostly written in Node.js, React, and a bit of Python. I don't claim to be an expert in these languages and was pretty unfamiliar with using them before I built these - so don't use this as a reference on how to write great code, rather I hope you enjoy the concepts behind the features.

Hope it can be of interest and potentially of help to someone out there!

Hey HN -

I made a bunch of GenAI projects in 2024: - 2 frontend React/Tailwind codegen projects (1 an agent, and 1 a really cool website builder) - 1 Node.js codegen AI agent - 1 ecommerce product placement image generator

I'm not working on these anymore (and the code and dependencies are probably far out of date now unfortunately - haven't looked at some of these for a year at this point).

But I thought I would just go ahead and make the code public and share it out there in case it can help anyone or inspire some new ideas.

Besides just sharing the code, if you click through to the detailed pages for each project you can see a demo video of each one showing how it works and read some of my notes about each one.

About the projects:

I really love the website builder, it's probably my favorite project I've done - so many cool details.

Other projects have some cool agent things going on, maybe some novel approaches to codegen in there - not sure.

The image generation one has a bunch of ComfyUI workflows that I spent a bunch of time on.

Happy to answer any questions here - although I'll probably have to review the code if anything is too in the weeds as I've forgotten a lot already.

Also the code is not very well-organized and there's some dead code and stuff floating around and things commented out or half-implemented at places. I didn't bother to clean it up too much besides making sure I'm not leaking any env secrets or API keys (if you spot one, let me know please!)

And last - these are mostly written in Node.js, React, and a bit of Python. I don't claim to be an expert in these languages and was pretty unfamiliar with using them before I built these - so don't use this as a reference on how to write great code, rather I hope you enjoy the concepts behind the features.

Hope it can be of interest and potentially of help to someone out there!

Yes, it is very good. I have built full websites complete with database storage and automated email sending all without having to touch the code at all.

I would say Lovable is the best performing that I've seen. I can build complete non-trivial sites without doing any coding at all.

Deep Research works reliably but I wish the agent would do some due diligence on its link selection and also today I uncovered where it misquoted a website which was very misleading.

Another good case is for rolling back a single bad commit from a batch that got merged into main at the same time.

Doing batch merges with a merge queue can speed up things if you have a ton of longer running end to end and integration tests. But then if a test fails you need to identify which commit out of the batch is causing it so you don’t reject the entire batch.

Now with AI test frameworks like stagehand it’s actually possible to write end to end tests after a bug appears that can be backwards compatible as long as changes to the dom are not too extreme. But things like broken selectors won’t be an issue.

I wrote about that here: https://qckfx.com/blog/ai-powered-stagehand-git-bisect-findi...

No, but happy to chat about it. My email is chris.wood@qckfx.com

In my experience, mandatory code reviews seldom work very well. Usually it's either stamping PRs or you always run the risk of someone just blocking stuff in an unreasonable manner.

For the most part, code reviews should be optional - if you want to get a review from someone, tag them on your PR and ask. If someone you didn't tag spots something and your PR landed, you can always figure it out and still make a fix.

I will give an exception to maybe super fragile parts of the code but ideally you can refactor/build tests/do something else that doesn't require blocking code review to land changes.

Thanks for the heads up - it's just a form to set a meeting right now.. If you're interested shoot me an email at chris.wood@qckfx.com.

The MCP server is useful, I built a demo of converting bug reports into Playwright tests using it - https://qckfx.com/demo

I’m building this because unable to repro bugs is one of the biggest reasons that bugs don’t get fixed. This leads to communication overhead as engineers and support teams go back and forth in a game of ping pong to try to determine who owns a bug.

I saw this first hand when I worked on operational excellence at Meta. Large companies can afford to hire teams to manually reproduce bugs before passing off to developers, but that option is not available to most companies.

It would be cool if you could build an autoencoder that took code in python or another less esoteric language and learned to output it into piet. Then you could do something like stable diffusion and get random algorithms.

This looks awesome, automated UI testing is so hard to get right but also very important. Great work so far!

Recently I stumbled upon Avida, a research project at Michigan State. They are exploring self-replicating & mutating computer programs as a way to learn more about natural evolution.

I can also see benefit in using this to develop an AI that is potentially more data/power efficient than something like GPT. In the same line of thinking, there is the potential for novel algorithm discoveries since evolution is based on chance mutations vs being constrained to human creativity.

Has there been any other similar research into this from an AI perspective? If not, why?

From this article: https://techcrunch.com/2023/11/16/codegen-raises-new-capital...

It sounds like the focus is on migrations, refactors, hiding stuff behind feature flags, etc. That's useful but it's less than what it sounded like from codegen's marketing materials which makes sense, they are selling the vision and not the current product.

If it's limited in handling complexity right now then it sounds like a nice feature for Linear. That also sounds like a good way to enter developer's workflows if you have some button on linear tasks to just automate it. That's not L5 engineer though and I bet it's expensive on the order of thousands of dollars per month for the OpenAI use.

Congrats to the team on the round, excited to see how it develops.

I think this is the obvious long term path for software development, but hard to know if it actually works since there's no way to try it. I also wonder if it has rough spots then how well will it be adopted vs tools like copilot that don't require such major change to developer workflows.