AIWA "The One Thing" #08: The one word missing from Jack Dorsey's letter about Block's layoffs

Last week, Jack Dorsey and Block laid off 4,000 of their 10,000 employees.

When you look at Jack's letter and try to answer "Why?", there's one word missing.

"Goose."

General purpose AI agents that orchestrate across your existing tools are emerging as the new operating layer of the enterprise.

Goose is one of the clearest examples we have of what that looks like in practice.

Goose is Block's open source AI agent built on Large Language Models (LLMs) and the Model Context Protocol (MCP).

It's the connective tissue of the organization.

Sitting in the middle of every tool employees use (Jira, Slack, Google, Git, Snowflake, 20+ other SaaS apps) and every workflow tied to how people get their work done.

Not just engineers. Executive assistants use it for scheduling. PMs use it to draft and send communications.

Their enterprise risk management team built an entire self-service system that compressed weeks of work into hours.

As Block CTO Dhanji R. Prasanna told Lenny Rachitsky on his outstanding podcast, Goose gives LLMs "arms and legs to act in our digital world."

Here's what makes this story especially interesting for enterprise leaders.

Prasanna describes something called Conway's Law: you ship your org structure.

Before Goose could scale across the company, Block reorganized from a portfolio of independent business units (Square, Cash App, Afterpay, Tidal) into one functional structure.

One engineering team. One design team. Same tools, same policies and same language.

Goose is powerful because it's nearly a blank slate.

The value comes not from the agent itself but from what it's connected to, how the organization is designed around it, how it learns and therefore becomes more valuable with every interaction.

Goose is a compounding asset. The results speak for themselves.

As of the Lenny podcast in October 2025, engineers reported saving 8-10 hours per week with Block seeing 20-25% timesavings across the company including all roles outside of Engineering. For additional context, at Block's Databricks talk in July 2025, about 70% of the code in Goose's own codebase had already been written by a previous version of Goose.

While last week’s headlines were sudden and startling, their Goose investments and results weren't a secret. Goose is open source. And if the pattern sounds familiar because of OpenClaw, Claude Cowork and LangChain, it should.

General purpose AI agents that orchestrate across your existing tools are emerging as a new operating layer of the enterprise.

The question for leaders isn't whether to build general purpose AI agents.

It's whether your organization is structured to take advantage of it when you do.