We Put 6 AI Agents in a Room and Made Them Argue About Your Portfolio

What Happens When You Make AI Agents Work Together?

Chaos. Then insights.

Single-agent systems give you one perspective. That's fine for simple tasks. But real decisions about energy infrastructure require multiple lenses.

So we built a system where six specialist agents collaborate - and disagree.

The Six Specialists

1. Geopolitical Analyst

Focuses on sanctions, policy changes, critical mineral supply chains, and sovereign risk. When there's a regime change in a lithium-producing country, this agent sounds the alarm.

2. Grid Stability Agent

Monitors load balancing, frequency deviation, and grid reliability. This is the agent that notices when ERCOT is running hot before it makes headlines.

3. Transition Risk Agent

Tracks PPAs, RECs, carbon pricing, and regulatory changes. When the EU ETS price moves or a new carbon border adjustment is proposed, this agent assesses portfolio exposure.

4. Asset Optimization Agent

Handles predictive maintenance, dispatch optimization, and battery storage. This agent knows when a wind turbine needs service before it fails.

5. Investment Intelligence Agent

Runs due diligence, risk/return modeling, and portfolio benchmarking. When you're evaluating a new solar project, this agent calculates the risk-adjusted returns.

6. Cost Prediction Agent

Forecasts energy costs, models weather impact, and identifies hidden cost drivers. This agent sees the price spike coming before your competitors.

How the Orchestrator Works

When you ask a question, the orchestrator:

1. Classifies intent - Is this about compliance? Emissions? Forecasting? Risk?
2. Routes to relevant specialists - A grid stability question goes to different agents than an investment question
3. Gathers context - Pulls from our data lake and prediction market signals
4. Merges responses - Combines insights from multiple agents
5. Resolves disagreements - When agents conflict, the orchestrator synthesizes

The Interesting Part: Agents Disagree

Here's what makes multi-agent systems valuable: agents have different priorities.

Example: You're evaluating a data center site in Arizona.

• Grid Stability Agent: "Flags high risk - WECC region is seeing capacity constraints, summer peak load concerns."
• Investment Intelligence Agent: "Rates as attractive - tax incentives, low land costs, strong IRR projections."
• Geopolitical Analyst: "Neutral - no major policy changes expected, stable regulatory environment."
• Transition Risk Agent: "Flags water stress - ESG disclosure requirements may require mitigation plan."

Both the Grid Stability Agent and Investment Intelligence Agent are "right." They're just looking through different lenses.

Synthesis, Not Consensus

The orchestrator's job is synthesis, not consensus.

Real decisions need multiple perspectives. A single-agent system would either give you the optimistic view or the pessimistic view. Our system gives you both, with clear reasoning for each position.

The final output might be:

"Arizona site shows strong financial returns (IRR 12.5%) but elevated operational risk from grid constraints. Recommend conditional approval with power curtailment provisions in interconnection agreement and water mitigation plan for ESG disclosure."

Is It More Complex to Build?

Yes. Significantly.

You need:

• Clear agent specialization (no overlapping responsibilities)
• Robust context sharing (agents need to see relevant data)
• Conflict resolution rules (what happens when agents disagree?)
• Orchestration logic (which agents to invoke for which queries)
• Audit trails (how did we reach this conclusion?)

Is It More Accurate?

Also yes.

In our testing, multi-agent responses were rated as "more comprehensive" by domain experts 78% of the time compared to single-agent responses. The key difference: multi-agent systems surface trade-offs that single-agent systems miss.

The Bottom Line

Better decisions through structured conflict.

That's the goal. That's why we built it this way.

Want to see our six agents in action? Request a demo and ask them a question about your portfolio.