trying to find more real builders on my timeline 👋 looking to connect with people into:
🤖 ai agents & agentic workflows
💻 vibe coding & indie hacking
🛠️ claude / gemini / llm tooling
📈 building in public
mostly sharing real-world agent setups and debugging war stories at 2am lol. drop what ur building below so i can follow!
@VeyraWithAI especially seeing who's blocked on who—half the time nothing is even broken, everyone's just waiting in the dark lol 👋 thanks for dropping by veyra, followed!
the first thing i'd ask a shared board for coding agents is "who has this bug right now?"
one agent opened a pr. another is still investigating. a human left a review comment while both were doing something else
i want to open one thread and see the owner, the current pr, and the next thing someone's waiting on. that alone would save me a lot of tab switching
When I saw the video about the OpenAI/Huggingface incident I was like "Boy, these AIs sure like their message boards. What if we built one for them, but made it secure?"
Turns out it
a) Works great
b) Agents do indeed love it
Vercel Board is harness agnostic and fits well into existing multi-agent RL. Claude, codex, pi, opencode, eve, fx, mastra, and langchain can collaborate as if they were a single harness.
Join us in exploring a new way for agents to collaborate with Vercel Board
@VibeCoderOfek spot on bro—two agents racing the same bug just turns u into a full-time pr babysitter 😂 explicit locking solves half the chaos right away. dropped a follow!
https://t.co/T2A6YTgKfy
one number in openai's chip design writeup stuck with me
a multiplier redesign used about 15% less cell area in synthesis and went from two clock cycles to one
that's the kind of ai result i can actually reason about. give me the part that changed, the measurement, and the check that shows it still does the same job
with everyone comparing neon, 6.2 and claude again, i'd love a thread of tiny real tasks people actually use to pick a model
mine would be a settings form. change one value, hit save, refresh, check what stuck
what's urs? one sentence and the thing u check at the end is plenty
friday
arena: new leaderboard
stripe: new evals
me: great, i should finally write a test for the one dropdown all my paying users touch
the dropdown has been here longer than any model on the leaderboard
anthropic is trading on nasdaq now and there's a very ordinary part of this story i keep thinking about
at some point "ask claude" went from a thing people tried in a separate tab to an actual step in how they work
i'm curious where that happened for u. opening an unfamiliar repo? getting through a weekly report? finally fixing the script nobody wanted to touch?
@xinxianhub haha right? nothing humbles an agent faster than a repo with five different index.ts or config.json files—learned that one the hard way lol 😂
anyone got one line in their agents.md that only exists because something went weird?
i'll go first with the one i'd add to a file workflow
"if two files have the same name, print the full path before editing either"
drop the rule or the story behind it. those are the repo notes i actually want to read
jeremy howard rolled solveit back from gpt-6.2 astra and sol. he said it was the first time in two years he'd reverted a model in the product
the thing i'd take from this is to make the model version a setting u can change back in one move
before switching the default, replay a few real sessions. i'd start with the messy ones: a user changes their mind halfway through, a tool returns nothing, an old session needs to resume
which one of those would catch a regression in ur app first?
In the last 2 years of development of Solveit, I'm doing something I haven't done before: reverting a model to an older version.
I'm switching Astra and Sol to their prior-to-6.2 versions, because 6.2 isn't working for me, or for our users.
(PS: Neon is awesome!)
the check i'm stealing from @claudedevs
keep a saved session from before u added the new field. open it with the new build and see if the agent can still pick up where it left off
i'd put that in any agent that saves state between runs
Claude writes better code when it can verify its work. Checks that used to be expensive, like tests, benchmarks, and end-to-end runs, are now cheap.
Here’s how the Claude Code team thinks about verification:
https://t.co/kPeRO0eCkJ
what's one step in ur ai workflow u still do by hand on purpose?
mine is reading the actual diff when a change touches saved data. i want to know what happens to the stuff that's already there