@ryolu_ I've had to drop cursor as im on wsl2 on windows and cursor has become incapable of attaching to the distro. vscode and windsurf always attach almost instantly. cursor is stuck in this loop. any help?
@owocki@devanshmehta agreed, but one could track/check if patterns emerge from those scenarios and limit the surface for bad actors who use those emergent vectors i guess
ah yeah, change being dependent on domain makes sense, thanks!
hey @devanshmehta@owocki when a project gets funded and later turns out to be a mistake, what was usually the thing you missed in the application? And the reverse: what do applications that turn out to be sleepers have in common?
@devanshmehta@owocki while I understand the poker analogy this is the reason I scoped it to applications. "Resulting" applies when you can't control the variables. Grant applications are the variables you can control. I guess I was asking "which pre-outcome signals predicted impact?"
Everyone in the world has to take a private vote by pressing a red or blue button. If more than 50% of people press the blue button, everyone survives. If less than 50% of people press the blue button, only people who pressed the red button survive. Which button would you press?
The whole point of shadcn/ui is I don't decide for you.
New: when you create a new project, look for the switch to turn on cursor: pointer for buttons.
Also available via the cli. npx shadcn init --pointer
@donnfelker@0xPolygon been building exactly this kind of stuff recently: agent harnesses, AI dev workflows, and a feedback-to-decision pipeline. AI tools are core to how i build, let's chat
https://t.co/aBNDoBIjUk
https://t.co/gWpTuxcBXM
just found out about @intelligenceco and I've been basically making some of their offerings. competitive analysis + blog scraping + email sending
maybe I should dust it off
learning a lot building a system that monitors competitors automatically. checks websites for price changes, scans news, watches social media. one weekly email with what changed and why it matters. if one source goes down you still get a report from the others
no UI (for now?) but its a multi-agent FastAPI pipeline: parallel collectors, pgvector dedup, LLM synthesis, retries, dead letters, structured logging, then it renders an executive HTML brief and sends it on a schedule.
Set targets, walk away, get the brief Monday morning.
actually having fun making this harness ruleset by figuring out the mental model. now to turn it into proper agent harnesses and commands and test in a project! @zakelfassi any tips on (agentic coding) harness testing?
biggest learnings so far:
+ task queueing via celery + redis
+ multi tenant considerations (ensuring data for one user doesnt leak in another user's request response)
+ @pydantic 's neat `pydantic_ai` package for model interactions
currently working on a system that takes your customer's feedback and gives you actionable insight.
ingestion pipeline and AI classification are working! sentiment, themes, urgency all auto-tagged.
you can also correct anything the AI got wrong and it learns from that.