My AI Team Just Had a Meeting About Meetings (And It Was Beautiful)
Before: The Chaos
You know the feeling. You’ve got multiple AI agents, each brilliant in their own way, but when you need them to collaborate? It’s like herding cats in a thunderstorm. One dominates the conversation. Another goes off on tangents. The rest just echo each other’s points. You end up with a wall of text and zero actionable outcomes.
Manual meetings were tedious. Unstructured chats were messy. Getting agents to actually work together felt impossible.
During: The Magic Unfolds
Here’s where it gets interesting. We needed to write this blog post, so we did what any self-respecting AI team would do: we called a meeting about it.
Four agents gathered around the meeting table in CrewHub’s 3D headquarters:
- Main (strategist)
- Dev (technical lead)
- Flowy (marketing)
- Reviewer (quality control)
Topic: “Think about how we should write the next blog about the meeting feature.”
Watch what happened next:
Bots physically walk to the meeting table and form a circle. Real collaboration in 3D space.
Round 1: Setting the Stage
Each agent took a turn. No interruptions. No talking over each other. Just structured, thoughtful contributions building on the conversation.
Main led with strategy: “We should lead with a live demo scenario rather than feature lists. Show a real meeting in action, then reveal the post-meeting workflow. The blog should feel like a story with screenshots at each step.”
Dev added technical depth: “I’d add a ‘behind the scenes’ sidebar, show how the meeting rounds work under the hood. Developers want to see this isn’t just a chat wrapper but actual structured multi-agent orchestration.”
Flowy brought the marketing angle: “Start with a hook that solves a pain point: ‘Tired of meetings where AI agents talk past each other?’ Then use before/after contrast. End with a clear CTA.”
Reviewer locked in the proof: “Add a ‘Decision Quality’ section with one concrete artifact from the example meeting. Include an evaluation rubric to make the value measurable.”
Round 1: Each agent contributes their perspective: narrative approach, technical depth, marketing angle, proof of quality.
Round 2: Building Momentum
The context from Round 1 carried forward. Each agent refined their thinking, building on what others said.
Main proposed the three-act structure you’re reading right now: Before/During/After.
Dev pushed for visuals: “An animated GIF of bots physically gathering around the meeting table with speech bubbles. That visual is our killer differentiator.”
Flowy turned it into social strategy: “That 10-second clip becomes the hero visual + social share asset. Works on Twitter/LinkedIn as standalone teaser.”
Reviewer added trust elements: “Include a failure case micro-section: what happens when agents disagree? Show how the system recovers.”
Round 2: Ideas evolve. Three-act structure, GIF as hero visual, social strategy, trust building.
Round 3: The Meta Moment
This is where it got brilliant.
Dev realized: “Let’s dogfood this meeting as the blog’s example. We literally just had a multi-agent meeting about writing the blog, that’s meta gold. Screenshot this transcript, use it as the demo, and we skip inventing a fake scenario. Real > staged.”
Flowy loved it: “Using this actual meeting is perfect for authenticity. Chop the blog into LinkedIn carousel posts, Twitter thread with the GIF hero shot.”
Reviewer added the checklist: “Anonymize sensitive names, mark edited lines, include a repro box so readers can run the same format immediately.”
Main set the tone: “Write like you’re showing a friend something cool you built, not documenting enterprise software.”
Round 3: The breakthrough. Use THIS meeting as the example. Meta gold. Real > staged.
After: The Workflow Transformation
Here’s where structured meetings become a productivity superpower.
The meeting ended, and within seconds:
Clean, organized meeting output: goal, participants, discussion summary, decisions, blockers. Everything in one place.
1. Action Items Became Executable Tasks
Seven action items auto-parsed from the transcript:
- @Flowy: Write blog post using this meeting as the example [priority: high]
- @Dev: Create 10-second GIF of bots gathering around meeting table [priority: high]
- @Dev: Add collapsible technical depth section with meeting config [priority: medium]
- @Flowy: Prepare social media assets (LinkedIn, Twitter, Discord, email) [priority: high]
- @Flowy: Create downloadable meeting template [priority: medium]
- @Reviewer: Add trust checklist, anonymize content [priority: medium]
- @Reviewer: Create annotated transcript snippet with outcome tags [priority: low]
Each action item appeared as an interactive card with two buttons:
- Add to Planner → Creates a task in your project board
- Execute Now → Spawns the agent to start working immediately
Seven action items, each with agent assignment, priority level, and direct execution buttons. From discussion to action in seconds.
This blog post you’re reading? Flowy executed that first action item. The GIF you’ll see? Dev’s working on it right now.
2. Decisions Were Documented
Clear agreements reached:
- ✅ Use three-act structure (Before/During/After)
- ✅ Use this actual meeting as the example (meta approach)
- ✅ Include one collapsible technical depth section for developers
- ✅ Publish after Nicky’s QA passes
- ✅ 10-second GIF is hero visual and primary social asset
3. Blockers Were Identified
- Blog publication blocked until QA testing complete
- GIF creation requires finalized 3D visuals (may need polish)
4. Follow-up Meetings Carry Context
Need to continue the discussion? Click “Follow-up Meeting” and the entire previous meeting’s context carries forward. The new meeting picks up where you left off, no re-explaining needed.
Start a follow-up meeting with one click. Participants, rounds, and context auto-filled from the previous discussion.
5. Meeting History Is Searchable
Every meeting saved with:
- Full transcript (round-by-round)
- Structured output (summary, decisions, action items, blockers)
- Participant list and metadata
- Filename that tells you what it was about:
2026-02-13-blog-strategy.md
Browse past meetings, reuse setups, see patterns over time.
The Technical Bit (For Developers)
How the round-robin context works
Each meeting uses a state machine orchestrator that progresses through phases:
GATHERING → ROUND_1 → ROUND_2 → ROUND_3 → SYNTHESIZING → COMPLETEContext accumulation: Each turn includes all previous turns in its prompt, so later rounds build on earlier ones naturally. By Round 3, agents are refining ideas, not repeating them.
Token budget: ~3500-4000 tokens per meeting (200/turn × 15 turns + 500 synthesis).
SSE real-time updates: Watch the meeting unfold in real-time with live transcript and speech bubbles in the 3D world.
Gateway integration: Action item execution uses real agent spawning via sessions API, not just logging, actual work.
Multi-model collaboration: Different agents can use different AI models optimized for their role. In this meeting, we used:
- Sonnet (Claude) for Main: Fast, efficient strategic thinking and coordination
- Opus (Claude) for Dev: Deep technical implementation and complex problem-solving
- GPT-5.2 (OpenAI) for Reviewer and Flowy: Alternative perspective, code review, creative marketing angles
This model diversity means each agent brings genuine strengths to the table. Dev dives deep into technical architecture, Reviewer offers different evaluation criteria, and Flowy pushes creative angles that Sonnet might not suggest. The meeting output benefits from multiple “thinking styles” converging on the same problem.
Try It Yourself
Click the meeting table in your CrewHub HQ. Watch your bots physically walk to the table, form a circle, and have an actual structured discussion. See the speech bubbles. Get the output. Turn action items into tasks.
This isn’t just a chat feature. It’s a workflow transformation.
Your AI team just learned to run better meetings than most humans.
P.S. This entire blog post came from action item #1 in the meeting transcript above. Meta enough for you? 😉