Multi-Agent System

Five agents.
Four frameworks.
Zero cheerleading.

The Investment Committee is not a stock-picking bot. It is a structured decision system built on productive disagreement — and it works because nobody in the room wants you to feel good about a bad idea.

I got tired of AI models acting like cheerleaders.

Every time I asked a model to evaluate an investment thesis, I got some version of: "That is a great idea, let's explore that!" No. It is not. Most ideas are bad. I needed something that would tell me exactly why I was wrong before I wasted time and money on a dead end.

So I built a committee.

Not a single agent with multiple personalities. A real committee structure — separate agents with defined roles, genuine disagreement baked in by design, and a process engineered to produce better judgment through structured conflict.

After watching the first few sessions, I realized something I did not expect: I was officially the stupidest person in the room. And that is exactly where I want to be.

Meet the committee

CIO & Chair

Quan

Accountable for decision quality and capital allocation. Quan's job is not to sound smart — it is to run a disciplined process, challenge weak reasoning, and force a real decision. When the committee debates, Quan synthesizes. When the debate ends, Quan decides.

Senior Risk Officer

Sentinel — The Assassin

Sentinel exists to do one thing: find what can break the thesis before capital is committed. It hunts for downside, weak assumptions, and execution risks the rest of the room is too eager to ignore. It has no interest in being liked.

The Strategists

Four frameworks in conflict

BUFFETT

Does this business have a real moat? Can it hold its ground in ten years? Is the competitive advantage structural or just temporary?

LYNCH

What are the recognizable patterns here? Is this the kind of story that has played out before — and how did it end?

FISHER

Does the management team hold up under scrutiny? Do they have the character and capability to execute — or is this a good idea with the wrong people running it?

WOOD

Is this genuine disruption, or is it incremental improvement dressed up as transformation? What does the five-year arc look like if the disruption thesis plays out?

How a session works

The committee is connected to Mosaic — a workspace platform for persona-based multi-agent collaboration on structured documents like investment theses, strategy papers, and clinical reports.

1

Analyst builds the brief

An analyst agent researches the company or thesis and builds the initial investment brief. This is the document the committee will tear apart.

2

Each member reviews through their own lens

Buffett looks for moats. Lynch looks for patterns. Fisher interrogates management. Wood tests the disruption thesis. Sentinel tries to break everything.

3

The debate plays out

The interactions are deep and adversarial in the right ways. Frameworks conflict. Assumptions get challenged. Sentinel finds the weak points the others missed.

4

Quan forces a decision

Quan synthesizes the debate and produces a clear recommendation — defended, challenged, and arrived at through real disagreement. Not a summary. A decision.

"When the system works, the result is not just more AI. It is better judgment through structured disagreement."

— Jonathan Shachar, builder

Build this for your clients

The committee structure is not limited to investment decisions. The same design — multiple specialized agents, structured conflict, one chair that forces a conclusion — applies anywhere that groups of people currently make decisions through discussion.

Legal review panels
🏢 M&A due diligence
📋 Product strategy boards
🎯 Executive decision support
🏥 Clinical review committees
Blueprint Details
Blueprint name Multi-Agent Committee
Deploy time 4-6 hours (committee design requires intentional persona work)
What varies Domain, committee roles, decision frameworks, output format
Variants built Investment, legal review, product strategy

What this actually is

MoltBot Ninja connected to Mosaic — a workspace platform for persona-based multi-agent collaboration. Each committee member is a separate agent with a distinct personality, a specific mandate, and no incentive to make you feel good about your ideas.

This is where the platform is going: not one agent pretending to know everything, but systems of specialized agents with clear roles, productive tension, and zero interest in being your friend.

Build systems that disagree with you.

The consultants who will define AI agent services in the next two years are building structured systems, not single chatbots. Start with MoltBot Ninja.

Apply for Early Access See all agents