Hooked Claude up to our Gaussian splat editor over WebMCP and asked it to delete everything except the airplane.
First run took ~30 min. Had it write down what it learned and ran it again: 4m38s.
Not efficient yet, but fun to watch.
People get high on abstraction too early. They want the system before they’ve earned the insight.
But the good abstractions are never designed. They’re discovered. You do the stupid manual thing enough times and the real bottleneck just emerges. Your initial agency might be driven by a hunch you had in the shower, but that moment won’t get you all the way to making something people want. The right way to make anything is forced on you by reality: what are the real jobs to be done? And what sequence?
This is why “do things that don’t scale” still hits, especially now when AI makes it trivially easy to scale things that probably shouldn’t be scaled yet. PG’s point was never about suffering. It was about contact. When you’re the one manually doing the loop, you see the edge cases. The weird user behavior. The failure modes nobody designed for. The hidden dependencies that only show up at 2am when some flow or intermediate step breaks in a way you didn’t anticipate. If you automate before you have that contact, you just scale your misunderstanding faster.
When the machines can help you vibe code perfection it gives you a false sense of power. I love that feeling as much as you do. But fuck perfection. Do it live. Be the loop.
Feel every friction point. Notice what’s actually true every single time versus what just looked true because you hadn’t seen enough cases yet. Formalize that. Build the recursive version. Then keep checking that your abstraction is still attached to real humans and their needs. Because reality drifts. Your users drift. The ground truth changes under you. You may think you understand but no plan survives contact with the real users and what they want. You find those body blows in analytics and user feedback and we call them the roadmap.
Humans left with not enough data hallucinate too. But just like the LLMs with enough data you unlock real transcendence. Real utility. Prosperity for humans in real life.
The abstraction is a tool, not a destination. The moment you forget that, you’re cooked.
I want one.
"when you’re deep in the flow, coding side-by-side with an AI, it stops feeling like a tool and starts feeling like a collaborator. And every collaborator deserves a presence on your desk."
@Azadux I tried this before, but didn’t have much luck, difference in exposure from different angles seems to make floaters. Did you restricted densification step somehow?
if "vibe coding", works, why don't they just vibe one up?
if it doesn't work, why would they want it?
Could it be the case where it doesn't work as well, but people will still be using them??
What's the angle here?
https://t.co/L4fA218bDw
Vibe Coding Tragedies and the Pit of Success
Heartbreaking to watch people lose months of work Vibe Coding in Cursor, an expert tool with zero guardrails.
At Replit, we obsess about getting customers to fall into the "Pit of Success," effortlessly adopting winning practices.
This term originally coined by Rico Mariani, a developer at Microsoft, and he defined it this way:
The Pit of Success: in stark contrast to a summit, a peak, or a journey across a desert to find victory through many trials and surprises, we want our customers to simply fall into winning practices by using our platform and frameworks. To the extent that we make it easy to get into trouble we fail.
This term was originally used in the context of programming languages, where if you use a language like C or C++ — especially as a non-expert — you can fall into traps like buffer overruns, memory leaks, and other potentially catastrophic issues.
In contrast, a garbage-collected language like Python, gives you zero control over memory management, but in return you get to avoid misery.
But this concept applies across the board to frameworks and platforms like Replit.
Our approach at Replit:
We give developers relatively few controls over the runtime and deployment environment and we make a lot of the choices for you. But in return you get to fall into practices that makes it impossible to lose work and protects your APIs and secrets by default.
For example, when using Replit Agent or Assistant, they will use Git, a version control system, whether you like it or not. They won't even work without it. That means it's almost impossible to lose work, and you get a lot of good side-effects like the ability to associate deployments with versions of the code.
Additionally, Replit manages the infrastructure for you, where we install the correct Linux packages, Python or JavaScript packages, and we do it in a transactional manner which reduces the possibilities of man-in-the-middle attacks and, like version control, makes state manageable.
We also manage cloud services for you like Database and Object Storage, where we take constant backups and we allow you to revert to earlier states in concert with version control. The Secrets and APIs for these servies are encrypted and stored securely, which makes it harder for hackers to abuse your services or steal your data.
Finally, we have higher-level services such as user authentication, where you can, with one checkbox, protect your app behind an auth layer that we manage that includes security features such as anti-botting.
Replit is not perfect, and it's still possible to find yourself in suboptimal or harmful situations, but you can always reach out to us or me directly for help.
Stay safe out there, and make sure to pick the right platform, especially if you don't know what you're doing.
I really try to avoid using open source projects that's backed by VCs. It's not really about the money itself.
I feel like the price is for being lazy to switch to something else. This one doesn't seem too bad though.
📥 Agent Inbox UI
Today, we're releasing the UI web app which powers our human-in-the-loop agent interactions.
The Agent Inbox UI is a generic app which can connect to, and interact with any LangGraph app that uses interrupt events for human-in-the-loop workflows
🧵