Multi-Agent - Consulting

DebateAI

A strategy engine where 100+ specialized agents argue over market-entry options, attack weak assumptions, score surviving strategies and assemble a campaign playbook.

Role Product Engineer Year 2025 Stack Next.js - LLM pipeline - Multi-agent orchestration - Three.js

A single model answer is too shallow for market strategy.

Market-entry strategy has competing incentives: growth, brand risk, cost, operations, channel fit and timing. A single prompt can summarize, but it does not create enough pressure between opposing views.

DebateAI turns the answer into a process. Agents take roles, argue, red-team each other, score evidence and promote only the strongest strategy into the final playbook.

The product workflow: brief to playbook.

The live showcase is one example run. The product itself is this workflow: intake a brief, create specialist positions, run debate rounds, score the tournament and produce a plan.

DebateAI new analysis screen with a filled company brief, competitor list, industry selector and selected target markets.

Brief intake

The strategy starts as structured product input, not a loose chat prompt.

The user enters company context, competitors, industry, markets, budget, debate depth and optional PDFs before the agent pipeline runs.

  • Markets are selected by region so comparisons stay explicit.
  • Budget and debate depth are part of the run configuration.
  • Documents enrich the evidence context without changing the workflow.
DebateAI analysis screen with phase tracker, live agent debate logs, completion state and decision audit.

Agent run

The interface exposes the workflow while agents debate.

Phase cards, live logs and the decision audit make the process inspectable: intelligence, market debate, campaign tournament and final validation.

  • Phase tracker shows which agent group is active or complete.
  • Logs reveal market advocacy, red-team pressure and validator checks.
  • Audit cards preserve winner, confidence, assumptions and unit-economics verdict.

A staged LLM pipeline with explicit scoring gates.

01

Brief intake

Normalize market, brand, budget and constraint inputs into a shared strategy context.

02

Agent debate

Role-specific agents produce arguments, counterarguments and revisions across debate rounds.

03

Tournament scoring

Strategies are ranked by market fit, risk, cost, evidence quality and upside.

04

Playbook assembly

The surviving strategy becomes a campaign plan with budget, channels, milestones and caveats.

What changed.

100+
specialized agents in the strategy process
4
phases: analysis, debate, tournament and playbook
8
markets compared in the example run
$1.3M
campaign playbook assembled from the winning strategy

More from the build.