@samecrowder@hwchase17@Hacubu It's complete 🎉 Prod switched today: ~31k threads copied into AI Chat → Production with their history, so no user lost a chat. Dev and local were easy. A built-in "move Agent Server to agent/environment" (or renamable agent ids) would have saved me most of a day.
@hwchase17@Hacubu
It's me, again 👋 More LangSmith feedback, this time on the new agents + environments (Production, Staging, Development, Local). Love the idea!
The migration turned each of our old tracing projects into its own agent: prod-chatbot, dev-chatbot, local 👇
@samecrowder@hwchase17@Hacubu Thanks Sam! Sending new traces to an environment worked right away 👍 The gap is existing Agent Servers: their agent is fixed at creation and agent ids can't be renamed, so moving old prod into AI Chat → Production meant a new server + copying ~31k threads via the API ourselves.
What would have saved the day 🙏
1️⃣ Rename an agent id (or add an alias)
2️⃣ A "Move to agent → environment" button for an Agent Server, threads and traces included
3️⃣ Or at least a server-side thread export/import between deployments
Option 2: a new agent. But an Agent Server's agent is set only when the server is created, so moving prod meant a new server and copying ~31k chat threads ourselves through the API, turn by turn, so editing old messages still works. Hours of scripts and testing.
@posthog@james406
PostHog spent a fortune on a blimp because you can't adblock the sky.
Unfortunately Dracula has been running ȚBlock Origin here since 1456. He invented the consent popup, nothing enters uninvited, and he blocks at the DNS level.
What I'd expect:
• If the cursor/focus is inside the document, Ctrl+F scopes to the document only
• Or a scope toggle in the search bar: Document / Chat / Both
Right now the search input lives in the document panel but doesn't belong to it.
Some friendly https://t.co/rNjqq9I4G4 UI/UX feedback @mkwng@joellewenstein 👇
Ctrl+F inside a document artifact also searches the chat, and the chat wins.
I ask Claude to draft something, it opens in the side panel.
👇 read next below
I had to hit next 10 times before the highlight finally landed in the document I was actually reading.
This only gets worse as the conversation grows. Longer chat = more chat matches queued ahead of the document ones.
It's me, again 👋 More LangSmith UI/UX feedback @hwchase17
Jumping between steps inside a trace is slow. Every click on a node in the tree = a visible wait with skeleton placeholders before the step data renders.
It gets a little bit worse,👇 read next below
once data arrives. Feels like a visual bug.
• Network tab open: opening ONE extra step fires 30-80+ requests (feedback, viewed_at, sessions, ls_user_id, stats, info — many repeated, each with its own preflight).
@james406@adamleithp Suggestion: only surface filters at the dashboard level when they affect every insight. Otherwise mark them visually (badge, scoped chip, "applies to 2 of 7 insights" hint) so stakeholders know what's actually being filtered.
Love the product otherwise 🦔🧡
Some friendly PostHog UI feedback @james406@adamleithp 👇
1️⃣ Open an insight from a dashboard and its filters don't show up at the top — but open the exact same insight from Product analytics and they're right there. Same insight, two different views of its own config.
👇 read
@james406@adamleithp 2️⃣ A dashboard shows all filters at the top, even when a filter only applies to one or two insights inside it.
For anyone on my team who isn't a daily PostHog user, this reads as "these filters apply to everything on this page" — which isn't true
👇 read