Workflow Automation · Read + Write

Virtual management meeting

Augment Rep · Medium Reason · High

The problem

Major decisions get rubber-stamped because the room agrees too easily. Real objections only surface in hallway conversations after the meeting — when it's too late.

The AI approach

Pre-meeting role-play. Multi-agent LLMs play the CFO, COO, CMO, CTO and Legal — each given the briefing + their actual mandate — debate for 3 rounds before the human meeting.

The outcome

~12 objections

Surfaces 8–15 objections + 3–4 risks the sponsor hadn't considered. Sponsor walks into the real meeting prepared.

Try itInput → Process → Output

Input — briefing + agenda
Decision briefQ3 board · sponsor: COO
DecisionCut SKU portfolio 4,200 → 3,000
DriverMargin pressure · ops complexity
TimingFY27Q1 · phased over 6 months
RevertsHard — supplier contract penalties
"Recommend approving. Will affect ~3% of revenue but lift gross margin ~80 bps."
Process — AI pipeline
1Parse briefing + agendaReadGenerative
2Spawn 5 role-agents (CFO/COO/CMO/CTO/Legal)RulesSymbolic
3Multi-turn debate (3 rounds)WriteGenerative
4Synthesise positions + risksWriteGenerative
Output — synthesised debate
Click Run demo to spawn role-agents and stress-test the proposal.
Agent positions 3 rounds
CFO
Approve — but ring-fence severance + supplier penalties.
COO
Approve, sponsor.
CMO
Concern — long-tail SKUs anchor 12 strategic accounts.
CTO
Defer — ERP migration overlaps, risk of bad data.
Legal
Concern — 4 contracts have 90-day termination penalty.
⚠ Top objections to address pre-meeting
  • CMO: 12 strategic-account SKUs — quantify revenue at risk before cutting.
  • CTO: ERP migration timing — phase the cut to land after Sept go-live.
  • Legal: 4 supplier contracts have 90-day termination clauses — net penalty ฿8.4M.
Recommendation: Approve with two conditions — quantify strategic-account exposure and shift go-live to Q2 to clear the ERP cutover.

Three AI types in this use case

SymbolicRole personas + meeting protocol (round-robin, time-box, topic gates); RACI rules per agent.
PredictiveSentiment / consensus scoring across rounds; flag agents who shifted position.
GenerativeMulti-agent LLM debate — each agent stays in role and applies its mandate; final synthesis writes the meeting brief.

The stack

  • Multi-agent · LangGraph / AutoGen
  • LLM per agent · Claude / GPT-4o
  • Persona library · per role
  • Synthesiser · vote + summary

When this works

  • Decision stake is high enough to justify ~30 min compute + review
  • Briefing is well-structured (agenda, options, constraints)
  • Role mandates are documented somewhere
  • Real meeting will follow within 24h

When it fails

  • Agents converge too fast — needs diversity-of-thought scaffolding
  • Persona drift — agent breaks character mid-debate
  • Treating the synthesis as a decision, not a stress-test
  • Confidentiality leaks — briefing data sent to model provider