**Important cautionary note: **
This writer is not qualified to give advice on anything - especially not on the subject matter.
Always consult your own consultants before following such advice.
**Yes — this is the moment. **
“Jack Dorsey and Block didn’t just experiment with AI. In February 2026 they made the leap: they cut roughly 4,000 people (about 40 % of the company) and publicly bet the entire business on becoming “intelligence-native.” They replaced layers of middle management and coordination with autonomous AI agents powered by their own open-source framework, Goose. Dorsey even published the manifesto “From Hierarchy to Intelligence.” The core idea is brutally simple and instantly actionable:
Stop using AI as a faster assistant.
Make autonomous AI agents the actual structural components of your company — the way org charts, departments, and workflows used to be built around people.
In an agent-first company, the “employees” are swarms of Goose-powered agents that plan, execute, coordinate, and improve themselves. Humans set vision, approve big moves, and handle the truly novel or ethical calls. Everything else runs on agents.How Goose Makes This Possible (The 5 Core Ideas Your Readers Need)Goose is not another chatbot or copilot. It is an open-source agent harness (now part of the Linux Foundation’s Agentic AI Foundation) that turns any strong LLM into a real worker that lives on your machine, connects to anything, and actually does the work. Here’s what makes it the perfect foundation for an entire company:
Agents = Teams (Subagents + Recipes)
One Goose instance can spawn unlimited specialized subagents on the fly. Each subagent gets its own focused tools, memory, and instructions.
You define repeatable workflows as simple YAML recipes (system prompt + tools + parameters). Run them once → they become reusable “standard operating procedures” that agents execute autonomously.
Result: Your entire org chart becomes a living swarm of agents instead of people in Slack threads.Everything Is a Tool (MCP = Model Context Protocol)
Goose talks to the real world through 70+ plug-and-play MCP extensions: your filesystem, GitHub, databases, email, Google Drive, Jira, Shopify, internal APIs — anything.
You (or one agent) can wrap your own company tools as MCP servers in minutes. Now every agent in the company can use them safely.Local-First + Model-Agnostic + Open
Runs on your laptop, desktop app, or servers. Use any LLM (local or cloud). Fully open-source (Apache 2.0). Fork it, brand it, create custom “distros” pre-loaded with your company’s tools and policies. No vendor lock-in.Safety & Human Oversight Built In
Plan Mode, permission levels (manual / semi-auto / auto), git snapshots for undo, approval rules, and red-teaming. Agents don’t run wild — they operate inside guardrails you control.Compound Intelligence
Agents talk to each other, self-correct, remember past lessons (episodic memory), and compound over time. Block’s own numbers: 25–40 % manual hours saved, engineers shipping ~40 % more code, and non-engineers now “vibe-coding” entire features.
How to Start Your Own Agent-First Company — Step-by-Step (So Readers Can Do It This Week)
Install Goose Everywhere (5 minutes)
Download the desktop app or CLI from goose-docs.ai. Connect your preferred LLM (Claude, GPT, Grok, local models — whatever). Add the first few MCP extensions you need (files, shell, GitHub, etc.).Map Your Company to Agent Structure
Product/Engineering → Main Goose + coding subagents + subrecipes for “build feature X”.
Operations/Finance → Recipe that pulls data → analyzes → forecasts → sends reports.
Sales/Marketing → Subagent swarm that researches leads, drafts outreach, updates CRM.
Customer Support → Dedicated Goose instance watching inboxes and auto-responding with tools.
Write one YAML recipe per core process. Test it in Plan Mode first.
Wire Your Tools Once (MCP Servers)
Wrap every internal system (CRM, accounting, inventory, etc.) as an MCP server. One afternoon’s work and every agent in the company can now use them securely. This is your new “company API” for intelligence.Launch the First Agent Team
Pick one high-leverage workflow (e.g., “take this messy CSV of sales data → clean it → build a Square store → send payment links”). Run it as a recipe with subagents. Watch it finish in minutes instead of days. Then turn that recipe into a permanent agent.Scale to Full Intelligence-Native Operation
Create a custom Goose distro for your whole team (pre-loaded with your MCPs, branding, and policies).
Replace meetings with agent hand-offs.
Use “flocks of Geese” (multiple coordinated instances) for complex projects.
Humans stay in the loop for strategy, exceptions, and creativity.Measure and Iterate
Track hours saved, output quality, and speed. Block did this and decided the old org chart was obsolete. You will too.
The Bottom LineA traditional company = humans directing tools.
An agent-first company = agents doing the work, humans directing intelligence.Goose gives you the exact harness Dorsey and Block used to make that switch. It’s free, open, local, and already powering real companies that moved faster than anyone expected.The wild rumpus isn’t coming.
It’s already here — and the first people to structure their entire company around agents will look like Block looks right now: leaner, faster, and impossible to catch.Install Goose today.
Write your first company recipe tomorrow.
Watch the old way of working become ancient history by next month.Your readers now have the map.
The only question left is: who starts building first? The island is yours. Let the rumpus begin.”
BEST OF FORTUNE TO ALL WOULD BE FOUNDERS OUT THERE
Be sure to let me know how the whole Wild Rumpus comes together.
The original information exchange is available here:
