Block · Goose general-purpose agent and G2 persistent workflow applications

A Evidence level

Block Investor Day 2025 – Full Transcript

Publisher: Block · Company investor material · Direct case-level source

Claim origin: Company management disclosure · Independent verification: No · Accessed: 2026-09-19

Evidence level measures whether a source can be located and reviewed; it does not mean vendor-reported claims were independently audited.

Deep case | Key delivery chain is substantially documented 12 / 12

Business context2 / 2

Transformation workflow2 / 2

Technical workflow2 / 2

Human roles & governance2 / 2

Measured outcomes2 / 2

Source traceability2 / 2

All six dimensions meet the current completeness threshold.

Goose addresses how an agent accesses data and takes action; G2 addresses how non-technical employees create durable applications. Together they form a path from tool connectivity to organization-wide agent adoption.

Business problem

Block wanted to extend AI beyond code assistance into HR, customer operations, legal, and other functions. A general chatbot could not safely read proprietary real-time data or take action, while non-technical employees could not wait for an engineering team to build every automation.

Solution

Block built and open-sourced Goose, an on-machine, model-agnostic agent that uses MCP to connect roughly 150 internal services and proprietary real-time data for both reading and action. It also built G2, a text-to-persistent-application environment where non-technical employees create “tiles” that run as ongoing asynchronous workflows. The same agent substrate is reused in customer products.

Technical architecture & production workflow

Step 01

Employee tasks, proprietary data, and real-time business systems

→

Step 02

On-machine, model-agnostic Goose agent

→

Step 03

MCP links roughly 150 services for read and action

→

Step 04

G2 converts natural language into persistent asynchronous application tiles

→

Step 05

Tool permissions, sandboxing, and employee confirmation constrain actions

→

Step 06

Internal feedback improves a shared agent substrate reused in customer products

Key technology & infrastructure components

GooseG2MCPOn-machine agent runtimeModel routingTool permissions and sandboxingPersistent asynchronous workflows

Human roles & accountability

Employees choose the task and tools and decide whether Goose should draft or complete a change. Business teams can create persistent workflows in G2. Tool permissions, sandboxing, job expectations, and business accountability constrain higher-risk actions.

FDE delivery actions

Connect agents to real-time proprietary data rather than only generic knowledge
Standardize internal read/write integrations through MCP
Generalize an engineering agent into a cross-functional substrate
Give non-technical employees a text-to-persistent-application interface
Track labor hours, support coverage, and engineering velocity separately

Reusable delivery patterns

A general enterprise agent needs a tool layer, not only a model layer
On-machine and model-agnostic design preserve data control and provider choice
Non-technical adoption improves when one-off chats become persistent workflows
Adoption, automation rate, and quality controls must be governed together

Business outcomes & delivery results

Block reports a 25% reduction in manual work hours across more than 75% of its employee base. AI handles 65% of Cash App support cases; more than 90% of code submissions are partially or fully AI-assisted; and median weekly code changes increased 30%.

Evidence boundaries & verification notes

The 25% reduction in manual work hours reflects Block’s broader AI tool portfolio and cannot be attributed solely to Goose or G2.
“More than 90% of code submissions” means partially or fully AI-assisted, not independently produced by AI.

Primary source: Block Investor Day 2025 – Full Transcript ↗ Additional sources: Block Open Source – Goose ↗ Additional sources: Goose documentation ↗