# US 2026/0310423 A1 · Hierarchical leader-quorum system for multi-agent model collaboration

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0310423 A1
- Document ID: US20260310423A1
- Status: Published application
- Kind: A1
- Application number: 19/098,794
- Filed: 2025-04-02
- Published: 2026-10-08
- Earliest claimed: 2025-04-02
- Assignees: Intuit Inc.
- Applicants: Intuit Inc.
- Inventors: Siddharth JAIN; Venkat Narayan VEDAM; Sivashanker THIRUCHITTAMPALAM
- Web page: https://patentgrep.com/patent/US20260310423A1

## CPC

- G06F 8/35: model driven
- G06F 8/36: Software reuse
- G06F 11/3495: {for systems}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Hierarchical leader-quorum system for multiple agent collaboration includes decomposing, by a large language model (LLM) of a central context management layer (CML), a software generation task specified in a natural language into multiple subtasks specified in a structured format. At least one subtask attribute of each corresponding subtask is matched to quorum data of a respective quorum, and transmitted, for each subtask assignment, the corresponding subtask to the respective quorum according to the subtask assignment. Further, a leader agent model of the respective quorum assigns the corresponding subtask to at least one agent model. The at least one agent model processes the corresponding subtask to generate software. The software generated by the quorums is combined. The software is presented.

## Claims

### Claim 1

A method comprising:
decomposing, by a large language model (LLM) of a central context management layer (CML), a software generation task specified in a natural language into a plurality of subtasks specified in a structured format;
matching at least one subtask attribute of each corresponding subtask of the plurality of subtasks to quorum data of a respective quorum of a plurality of quorums to obtain a plurality of subtask assignments;
transmitting, for each subtask assignment of the plurality of subtask assignments, the corresponding subtask to the respective quorum according to the subtask assignment;
assigning, by a leader agent model of the respective quorum, the corresponding subtask to at least one agent model of a plurality of agent models;
processing, by the at least one agent model, the corresponding subtask to generate software;
combining, by the central CML, the software generated by the plurality of quorums; and
presenting, after combining the software, the software.

### Claim 2 (depends on 1)

The method of claim 1, wherein the at least one agent model is a specialized agent model specifically trained to generate software for a subtask type of the corresponding subtask.

### Claim 3 (depends on 1)

The method of claim 1, further comprising:
receiving, via a graphical user interface, a task input describing a software generation task in a natural language format.

### Claim 4 (depends on 1)

The method of claim 1, further comprising:
constructing a directed acyclic graph (DAG) from the structured format;
annotating the plurality of subtasks in the DAG with the plurality of subtask attributes,
wherein transmitting the corresponding subtask to the plurality of quorums is performed in an order defined by the DAG.

### Claim 5 (depends on 1)

The method of claim 1, further comprising:
decomposing, by the leader agent model, the corresponding subtask into a second plurality of subtasks;
assigning each subtask of the second plurality of subtasks to the plurality of agent models associated with the quorum;
receiving, from the plurality of agent models associated with the quorum, the software; and
transmitting the software to the central CML.

### Claim 6 (depends on 5)

The method of claim 5, further comprising:
aggregating the software from the plurality of agent models according to performance metrics of the plurality of agent models.

### Claim 7 (depends on 6)

The method of claim 6, wherein aggregating the software comprises:
determining a difference exists between software of a first agent model of the plurality of agent models and a second agent model of the plurality of agent models;
selecting the software of the first agent model or the second agent model based on a performance metric of the first agent model and the second agent model with respect to the corresponding subtask; and
using the selected software as the software for the corresponding subtask.

### Claim 8 (depends on 1)

The method of claim 1, further comprising:
electing, from the plurality of agent models in the quorum, the leader agent model according to a plurality of leadership metrics gathered from the plurality of agent models.

### Claim 9 (depends on 1)

The method of claim 1, further comprising:
monitoring the generation of the software by the plurality of agent models; and
updating a plurality of performance metrics based on the monitoring to generate updated performance metrics; and
using the updated performance metrics to select agents for a second software generation task.

### Claim 10 (depends on 1)

The method of claim 1, wherein the subtask attribute defines an expertise required to complete the corresponding subtask, and wherein the quorum data defines an expertise of the quorum.

### Claim 11

A system comprising:
at least one storage device; and
at least one computer processor for executing instructions stored on the at least one storage device for causing the system to perform operations comprising:
decomposing, by a large language model (LLM) of a central context management layer (CML), a software generation task specified in a natural language into a plurality of subtasks specified in a structured format,
matching at least one subtask attribute of each corresponding subtask of the plurality of subtasks to quorum data of a respective quorum of a plurality of quorums to obtain a plurality of subtask assignments,
transmitting, for each subtask assignment of the plurality of subtask assignments, the corresponding subtask to the respective quorum according to the subtask assignment,
assigning, by a leader agent model of the respective quorum, the corresponding subtask to at least one agent model of a plurality of agent models,
processing, by the at least one agent model, the corresponding subtask to generate software,
combining, by the central CML, the software generated by the plurality of quorums, and
presenting, after combining the software, the software.

### Claim 12 (depends on 11)

The system of claim 11, wherein the at least one agent model is a specialized agent model specifically trained to generate software for a subtask type of the corresponding subtask.

### Claim 13 (depends on 11)

The system of claim 11, wherein the operations further comprise:
receiving, via a graphical user interface, a task input describing a software generation task in a natural language format.

### Claim 14 (depends on 11)

The system of claim 11, wherein the operations further comprise:
constructing a directed acyclic graph (DAG) from the structured format;
annotating the plurality of subtasks in the DAG with the plurality of subtask attributes,
wherein transmitting the corresponding subtask to the plurality of quorums is performed in an order defined by the DAG.

### Claim 15 (depends on 11)

The system of claim 11, wherein the operations further comprise:
decomposing, by the leader agent model, the corresponding subtask into a second plurality of subtasks;
assigning each subtask of the second plurality of subtasks to the plurality of agent models associated with the quorum;
receiving, from the plurality of agent models associated with the quorum, the software; and
transmitting the software to the central CML.

### Claim 16 (depends on 15)

The system of claim 15, wherein the operations further comprise:
aggregating the software from the plurality of agent models according to performance metrics of the plurality of agent models.

### Claim 17 (depends on 16)

The system of claim 16, wherein aggregating the software comprises:
determining a difference exists between software of a first agent model of the plurality of agent models and a second agent model of the plurality of agent models;
selecting the software of the first agent model or the second agent model based on a performance metric of the first agent model and the second agent model with respect to the corresponding subtask; and
using the selected software as the software for the corresponding subtask.

### Claim 18 (depends on 11)

The system of claim 11, wherein the operations further comprise:
electing, from the plurality of agent models in the quorum, the leader agent model according to a plurality of leadership metrics gathered from the plurality of agent models.

### Claim 19 (depends on 11)

The system of claim 11, wherein the operations further comprise:
monitoring the generation of the software by the plurality of agent models; and
updating a plurality of performance metrics based on the monitoring to generate updated performance metrics; and
using the updated performance metrics to select agents for a second software generation task.

### Claim 20

A method comprising:
decomposing, by a large language model (LLM) of a central context management layer (CML), a software generation task specified in a natural language into a plurality of subtasks specified in a structured format;
matching at least one subtask attribute of each corresponding subtask of the plurality of subtasks to quorum data of a respective quorum of a plurality of quorums to obtain a plurality of subtask assignments;
transmitting, for each subtask assignment of the plurality of subtask assignments, the corresponding subtask to the respective quorum according to the subtask assignment;
assigning, by a leader agent model of the respective quorum, the corresponding subtask to at least one agent model of a plurality of agent models;
processing, by the at least one agent model, the corresponding subtask to generate software;
combining, by the central CML, the software generated by the plurality of quorums; and
executing, after combining the software, the software.
