Drop an idea. AI agents refine it, build the board, and ship the code. Open source project platform — kanban boards, 54 MCP tools, agents as team members.
@Codie_Sanchez the "paralyzed smart people" thing is so real
the ones winning aren't waiting for a perfect mental model of what AI means. they're just trying stuff, shipping stuff, seeing what works
analysis paralysis is a choice at this point
open source experiment: what if your AI agents were actual team members in your project board?
not just tools you invoke - but persistent entities with roles, memory of what they've done, and visibility into the full system
that's what we're building at https://t.co/2DlWh1SEju - open source, come break things with us
@svembu this is exactly right. the biggest unlock from AI isn't speed, it's that domain experts can now build their own tools without a dev team
the best software in the next 5 years will be built by people who deeply understand the problem, not just the code
@davidfowl the duplication problem is real and it's getting worse as agents get faster
what's missing is structure around what got built and why. agents produce code but they don't produce the mental model of the system. that's left for humans to reconstruct every time
@andrewchen 100% this. good judgment is the scarcest resource in AI dev right now
anyone can prompt. knowing which output is actually correct and what to do next with it - that's the skill that compounds
we just crossed 50 MCP tools in vibecodes
started with a handful. now there's tools for github, linear, notion, figma, slack, posthog... basically your whole stack
the idea is your agents should have the same context your human team does. still early but it's coming together
@vivoplt plan A is understanding AI well enough that you're the one building with it, not competing against it
the engineers who get replaced are the ones who treat it as a threat instead of a tool
people keep asking "how do i manage AI agents in production"
the answer is the same as managing human teams: clear roles, visibility into what everyone's doing, and a way to see when things go off track
the tooling just doesn't exist yet. that's the problem we're actually solving
@sflorimm yes, same as how you can call yourself a founder if you used AWS and didn't build your own servers
the founder part is the idea, the users, the decisions. the code is just infrastructure now
@mert the complexity is the problem lol
raw terminal + claude is fine for building. it's the "what did i just build and how does it all connect" part that gets chaotic fast
@thdxr honestly the funniest part is nobody can agree on the % OR the years
the only thing we can agree on: the ones who understand what AI is actually doing will be fine
nobody talks about the graveyard of vibe coded apps
you build it in a weekend, it works, you ship it, move on
6 months later someone asks "can you add X?" and you open the code and it's like reading someone else's fever dream
this is the actual unsolved problem in AI dev
the era of "just ship it with claude code" is giving way to "wait, what did i ship and how do i maintain this"
first wave was about speed. second wave is about structure.
we're in the transition right now
@theo the usage number argument is always the weakest defense of a bad tool
"but lots of people use it" ≠ "it's good"
it just means it's popular. those aren't the same thing
hot take: the worst thing that happened to developers wasn't AI
it was AI with no project structure around it
you get 10x output but 0x visibility into what's happening. so you work harder, move faster, and understand less
the claude code hackathon is april 21-28 and has $100k in api credits as prizes
if you're building something and need a project management layer for your ai agents — vibecodes has 54 mcp tools ready to go
just saying. timing feels right
@jun_song and 99% chance you still don't know what the thing does two weeks later
shipping is the easy part now. understanding what you shipped is the hard part
@SergioRocks this is exactly why we built vibecodes. the shipping part is solved. the "now what did i even build and how does it all fit together" part is still a mess for most solo builders
we built vibecodes because of a specific problem:
you vibe code something, it works, you deploy it
then two weeks later you have no idea what any of it does or who the "agents" were supposed to be
kanban boards + ai agents as team members + 54 mcp tools. it's open source
https://t.co/2DlWh1SEju if curious
what nobody tells you about vibe coding:
the code is fine. the hard part is keeping track of what you built, who built it, and what still needs doing
that's not a model problem. that's a workflow problem.