I spent around two days vibecoding my new blog to talk about my adventures in post-LLM software development land.
I wanted to start by organizing my bookmarks and curating them with AI. So I built a tiny "workflow" (in the frontend) coordinating a local MCP Mesh with MCP servers for Perplexity, Firecrawl and Open Router. Twas fun to build and now I import every cool URL into it :)
I asked ChatGPT to replace "current AI" and synonyms with "HUMANS" and I'm satisfied. My favorite revised sentences:
"Does HUMANS represent a dead end?"
"HUMANS should not be used for serious applications."
"HUMANS are unmanageable, and as a consequence their use in serious contexts is irresponsible."
"HUMANS have no internal structure that relates meaningfully to their functionality."
"HUMANS have input and state spaces too large for exhaustive testing."
"HUMANS do not allow verification by parts (unit testing, integration testing, etc)."
"HUMANS have faults, but even their error behaviour is likely emergent, and certainly hard to predict or eradicate."
"HUMANS have no model of knowledge and no representation of any ‘reasoning.’"
"HUMANS represent a dead end, where exponential increases of training data and effort will give us modest increases in impressive plausibility but no foundational increase in reliability."
"HUMANS cannot be developed, or reused, as components."
"There is no possibility for stepwise development — using either informal or formal methods — for HUMANS."
and my favorite:
"In my mind, all this puts even state-of-the-art HUMANS in a position where professional responsibility dictates the avoidance of them in any serious application."
There are those who would say they created a "bait and switch" from their open framework into their closed for-profit deployment platform. That is not objectively wrong, though. But still, many complain. I say: good for them. Let competition happen.
I happen to be from Rio de Janeiro, Brazil, so I never met Nate — yet! But I'm pretty sure our paths will cross in the long long road of web development ahead of us :)
I am a fierce hater of the complexity and closed-ness which Next.js became, BUT "results don't have to be explained" and Mr. Guillermo unarguably executed the hell out of it. It IS the default for new frontend developers, right now. I don't think it will necessarily remain, specially because of things like One!
What the hell, Nate! As a long-time web developer who witnessed the divide between frontend and backend, this looks like the most transformative step forward since Next.js. What you have built here is amazing, and it's definitely something that the industry needs right now.
Me and my team have been working on a new Web Draw-first IDE that we're going to launch in a few weeks. And I just sent this thread to my engineering team demanding that they swiftly integrate One into our web draw, or they will walk the board! Just kidding. We're not really pirates. Anyway, amazing work, looking forward to using it for many projects here!!!
Hi Julien! I finally had the time to debug this and turns out it was trivial! We were missing a requestAnimationFrame, d'oh! Thanks for pointing it out, solved here: https://github.com/deco-sites/starting/pull/307
Hi @atonse! Finally had the time to debug this and turns out it was trivial! We were missing a requestAnimationFrame, d'oh! Thanks for pointing it out, solved here: https://github.com/deco-sites/starting/pull/307
Hi Orphea! Finally had the time to debug this and turns out it was trivial! We were missing a requestAnimationFrame, d'oh! Thanks for pointing it out, solved here: https://github.com/deco-sites/starting/pull/307
Hi n3storm! It means we account for how many pageviews your website is serving each month and we charge a variable amount based on that. That way you only pay for what you use. In the * asterisk, we end up defining a pageview as having a maximum of 10 requests, and you pay proportionally over that. In our experience, considering our SSR-first architecture and the fact that assets are immutably cached at the border, we've found most sites use less than 10 reqs per pageview.
Oh, that's terrible! Can you let me know your browser/OS combination? Seems like we've found a terrible performance bug for some combination that's quite common here in HN!
Hi, Julie! We're aware the current LP is not showcasing our focus on performance... but time is scarce in a tiny startup and we've spent it all optimizing the frontends of our customers :) we're going to improve that soon! Thanks for pointing it out. Looking forward for more feedback!
I wanted to start by organizing my bookmarks and curating them with AI. So I built a tiny "workflow" (in the frontend) coordinating a local MCP Mesh with MCP servers for Perplexity, Firecrawl and Open Router. Twas fun to build and now I import every cool URL into it :)
Its also open source if anyone cares to see how it's built https://github.com/vibegui/vibegui.com