greenfish6

Powerful. And written so well, I thought it was real at first, I even forgot about the "this is a short story at the top" and googled Aaron's different betting websites

It's also very representative of the HN comment section vibes overall.

But such is normal with normies providing startup critiques, if you just look at every announcement and say "this will fail", you'd be right 99% of the time.

oh, the point was actually the guessing game half a scroll down to see if you could determine which of the comments is AI vs human written

An agentic marketplace where people can buy business services from inside their claude code: https://talkshi.com

I've not found anything useful on x402 that is only available on x402. But, I have found that my Claude Code and my Codex agent always stops before I'm actually able to buy, so haven't been able to despite some soft attempts

Now that I think about it, it's very easy to point to reverse jevon's paradoxes. Regulations around building housing, large org bloat + processes, when the stupid form fill that i need to fill out gets too long on some website...

one of the things that early founders don't understand about acquisition though is that getting bought out in an acquihire usually precisely involves working for somebody else

You'd be surprised at how many startup founders don't calculate the probabilities of different outcomes and are not at all aware of said high school guidance counselor advice

technically, if you just told claude your credit card number and turned on computer use, then it should be able to buy all of these things, although probably slower than if you just did it yourself

I'm interested in this, I think this could connect to some things I'm doing

The biggest shift in the economy from AI is going to be who can figure out how to start up the agentic purchasing loop.

Similar to the fact that you didn't need a website to do biz in 1994, but you did need one do more business in 2004.

Most of the efforts so far like x402 I think are missing the point, because it is already super easy for an agent to sign up for a service and make an api key automatically. Capturing & executing business services via api, like perhaps consulting, accounting, event planning, through those entity's specialized agents is the interesting part imo

i mean yea

yes, so the thought experiment here is, how far can agents (owned by different entities) get in being productive without a human stepping in

How much can agents accomplish by coding with each other with 0 humans in the loop?

something i really like from tryin git out over the last 10 minutes is that the main agent will continue talking to you while other agents are working, so you don't have to queue a message

yea but i feel like we are over the hill on benchmaxxing, many times a model has beaten anthropic on a specific bench, but the 'feel' is that it is still not as good at coding

i would have to imagine the gastown design isn't optimal though? why 8, and why does there need to multiple hops of agent communications before two arbitrary agents communicate with each other as opposed to single shared filespace?

Excited to try this out. I've seen a lot of working systems on my own computer that share files to talk between different Claude Code agents and I think this could work similarly to that.

(i thought gas town was satire? people in comments here seem to be saying that gas town also had multi-agent file sharing for work tracking)

When i went it was blank? Lol i think someone made claude delete the site?

I use Willow AI, which I think is pretty good

The first example has a miscalculation; if you invest 1k and the EV is 900, then your choice has negative ROI, not positive.

One of my most anticipated tools is context truncation, happy to see them adopt certain practices that were previously only on the fringes like the ideas from https://ai.pydantic.dev/message-history/#summarize-old-messa...

pelican on a bicycle benchmark probably getting saturated... especially as it's become a popular way to demonstrate model ability quickly

As the rate of model improvement appears to slow, the first reactions seem to be getting worse and worse, as it takes more time to assess the model's quality and understand the nuances & subtler improvements

I wonder what Ankit's living in Boston means for YC's SF/bay area thesis for startup survival, which I thought they were very passionate about

Yes, but if you look in the rate limit notes, the rate limit is 500k tokens / minite for tier 4, which we are on. Given how stingy anthropic has been with rate limit increases, this is for very few people right now

I use claude code in a staging env also with full bash permissions via the open source github actions (I think this is much safer and also what you're supposed to do). The github issues and PR integrations make it really organized

Benchmarks don't tell you as much as the actual coding vibe though

Modern Realty (YC S24) | Founding AI Engineer | San Francisco, CA | Full-time

We're building an AI-powered real estate brokerage that allows users to buy homes without a traditional realtor. Our platform automates agent workflows including messaging, deal management, and customer interaction.

Tech stack: TypeScript, AI/LLM integrations with focus on tool use

Responsibilities: - Design and develop AI-driven workflows for messaging and deal automation - Build robust backend services using TypeScript - Craft effective GPT prompts and fine-tune AI responses - Collaborate closely with CEO/CTO on key platform features

Seeking someone who: - Uses AI to ship code fast - Has shipped consumer-facing features - Has good grasp of existing AI models

Compensation: $100K-$150K, 0.5%-2.5% equity

Apply / JD: https://www.workatastartup.com/jobs/66546

Worked for me once so far for a single github issue which had a solution that involved a read and write to my supabase, seemed to have a big upgrade from the codeact 2.1