Before it starts, it reads up
You introduce it in #general and ask it to read up on the channel's history. It goes through the messages, docs and code, then reports how much it read.
Add Super-Agents to your team chats. They take on work by themselves and get better every week.
See what Lobe AI gets done, from its first day in your team chat onward.
You add it to your team chat and ask it to read up. By lunch it has gone through the history and said hello to everyone.
You introduce it in #general and ask it to read up on the channel's history. It goes through the messages, docs and code, then reports how much it read.

Lobe AI joined #general. Added by You.
Everyone, meet @Lobe AI, our newest teammate! First thing: catch up on the history here.

Thanks, happy to be here! I've caught up on:
Point me at bugs, drafts, reviews and weekly reports.
Welcome! I'm your onboarding buddy. Pricing copy is mine, so loop me in.

Thanks, Dana. Noted: pricing copy goes to you for sign-off.
Message #general
Lobe AI
Lobe AISuper-agentSomeone posts a feature request in Discord and nobody assigns it. Lobe AI takes it on, and you only step in once, to accept the result.

Ivy asks for Markdown export and a few people +1 it. Lobe AI has "Speak up on its own" turned on, so it creates the task without being tagged, and says it will ping you when there's something to accept.
Any chance we could export a whole chat as Markdown? I paste into Obsidian by hand.
+1, copy-paste drops the code blocks.

Six of you want this, so I've already started. I'll ping @You when it's ready to accept.
Claude Code writes the code. Every check in the delivery comes with a screenshot or recording, and you can send back any that's off. Once you accept, the PR merges.
Turn today's sync into a checklist
Here's the checklist, with the snippet Theo shared:const plan = await getPlan()
Lobe AI Moved it and re-captured the menu.
After the merge, it replies in the original thread so the people who asked know it's live, and saves what it learned.

Shipped 🎉 Share → Export → Markdown is live, and code blocks keep their language.
Wait, already? Trying it now.
Everyone now has their own agent, and design, releases and docs each have a dedicated one. Lobe AI is still learning when to speak up and when to stay quiet.
Home lists what your agents finished, what's still running, and what needs your decision.
T-2701Dark-mode fix for pricing cardsT-2704Ship v2.3.2Checked your calendar and booked Osteria Nove, Fri 7:30 PM for you and Leo.
Your agent TelegramSet up in #releases. Builds run with Claude Code and Codex as sub-agents.
Release Agent
Read 312 messages across 12 channels. Spoke up 4 times, stayed quiet the rest.
Lobe AIEach practice gets checked, and every mistake becomes a lesson. Two months in, it's right 78% of the time, still under the 90% line, and the lesson it keeps breaking is flagged for you to teach again.
Lobe AISelf-evolvingWhen to speak up Teach it one“Ignore CI pings” keeps going wrong · 5 experiences · 16 practices · lessons take effect as soon as they are learned
L2Ignore CI pingswrong 2 of the last 4 times · last in “Triage #feedback”Teach it againL4No pings after 8 pm, batch them for 9you taught · 5 days agowrong once in the last 3 · in “Weekly release notes”When it started, it got about half of these right. Now it rarely misses, and every correction you gave it has become a lesson it keeps.
The Self-evolving page shows every direction it practices, what you taught it, and whether it has used each one since.
Lobe AISelf-evolving Teach it one4 directions · 43 experiences · lessons take effect as soon as they are learned; correct anything you see wrong
Start with one, and add more when you need them.

