starting from scratch.
CS student building AI products — agents, RAG, full-stack.
no existing audience, no product, no roadmap.
building everything in public. follow if you want to see how it actually goes.
@CompleteSkeptic Hey Diogo, I'm thrilled to see this, congratulations. I want to experiment with Jev. While there's plenty online, I'd like to create something unique and dive deep. Could you suggest a project I can build today and share on X?
I've seen many posts about building things with Jev. I'm interested in creating something and experimenting with it. I have my account, but I'm not sure what to build. What suggestions do you have?
Hey, something is coming from my side. This isn’t a product I’m trying to sell or make for profit; I simply felt it should exist, so I’m taking the initiative to build it for India as an Indian. I’ll share updates soon 🇮🇳
if agent checkout keeps eating the funnel, what's the first identity signal you'd actually fight to keep: email, device id, something else? curious what people building on this are prioritizing right now.
amazon sued perplexity over the comet agent and won an injunction blocking it from logged-in areas. the argument: it disguised automated activity as human browsing. cfaa and tos, not robots.txt. what counts as "a person browsing" just moved into a courtroom.
agentic checkout collapses browse-compare-decide-buy into one interaction. no cookie consent, no email capture, no account creation step. every identity checkpoint attribution has relied on for 15 years just isn't there anymore.
real question for anyone running ai agents against production data: are you logging every tool call they make, or just the ones that touch something destructive?
most bot detection is behavioral: mouse jitter, dwell time, hesitation before submit. that logic assumed only humans could produce it naturally. agentic browsers are now generating the same signals on purpose. the fraud model didn't break, the assumption underneath it did.
I'm also seeing this pattern for a few days in my feed. i can easily understand that they are ai generated and they are purely for reply farming and some followers farming also. i don't know what value it creates in the long term. followers are well needed but they are not going to convert what's the use if we can't provide any value,
pixel-based tracking assumes a browser renders the page, fires the tag, loads the dom the way a human's would. agentic browsers don't reliably do any of that. half the analytics stack built for the last 15 years is quietly going blind to a growing slice of traffic.
@keelenai the "tests must fail before implementation" part is the one people skip. how do you actually enforce that mechanically, a hook that checks the test ran red first, or is it just a review step?
teams standardizing on one coding agent are now writing real guardrails for it — repo access, budget per run, review before merge.
what does your actual guardrail setup look like in prod? mine's basically "don't touch main and pray."