The Trillion-dollar Terraform and the Last Job on Earth

$5T of value is going to be created over the next 7 years by a new kind of company with a new kind of role for everyone that works there.

There are only a handful of these companies today, and I had dinner with some of their founders a few weeks ago.

Between courses, one founder shared what he'd told his team at a recent all-hands. It perfectly captures what's happening:

"You aren't smarter than AI anymore. What that means is your job is no longer to do the work or even make decisions. Your job is to give the AI as much context as possible so it can make the decisions and do the work."

Surprisingly, about half the table had come to the same conclusion and restructured their entire companies around it. Everyone had built what they called a "company brain" — a single place aggregating all their context so agents had everything they needed to run the business.

And it produced incredible results. All of them were operating at a mind-boggling speed and scale — zero to tens of millions in revenue, in months, with tiny teams.

"Probably something" I thought, as I shoved more dumplings in my mouth.

The Agentic Micro Company

The writing has been on the wall for our 2,000-year-old model of company-building.

As Jack Dorsey and Sequoia's Roelof Botha pointed out in their recent essay "From Hierarchy to Intelligence," the dominant corporate structure with massive, specialized teams and layers of management was designed by the Romans to disseminate information in a world constrained by the limitations of the human mind.

But now that we have digital minds that can house massive amounts of context and execute a wide range of tasks, this model just slows things down and falls apart.

Dorsey announced a 40% layoff and restructuring of Block into tiny teams using agents to handle massive scope. It's the same exact model the founders I met at dinner had.

I call these "Agentic Micro Companies (AMCs)":

• Tiny teams of generalist humans
• Massive teams of agents
• Create custom software
• Powered by a singular company brain

Think of it as Founder Mode on steroids.

The Company Brain

Context is king, but for the most part, the emperor has had no clothes.

As has been proven in countless studies, even small models outperform larger models when they have better context. The more emails, meeting notes, and documents you feed an agent, the more powerful it becomes.

We've seen this ourselves. The Micro agent outperforms every other agent we tested (including fully loaded Claude Code, Codex, OpenClaw, & Hermes agents) for most queries, simply because it has an advanced system for storing and retrieving memory. Building the world's best agent memory sounds impressive. Honestly, the bar has been pretty low.

Today's agents are basically quantum physics PhDs with severe amnesia.

It wasn't until Karpathy's essay on knowledge graphs (March 2026) that the industry truly woke up to the importance of unified memory systems, or "brains" as some are calling them. The open source ecosystem has exploded with different approaches, from Graphiti to Mem0, but at the end of the day, the best ones ultimately all approximate the functions of the human brain: long term memory, episodic, short term, etc.

AMCs deploy a single brain for their entire company that can be used as a backend for custom software they build, for quickly looking up information, and to be the hive mind for hundreds of agents that run their businesses every day.

The Context Farmer

As model intelligence grows, so does the human role:

Model Intelligence	Human Role	What that looks like
Lowest	Individual contributor	You do the work yourself
Low	Micromanager	You tell AI exactly what to do, step by step
Medium	Director	You set goals and review AI output
High	Strategist	You define the problems worth solving
Highest	Context Farmer	You tend the brain. AI runs the business.

The final job humans will have is Context Farmer.

Farmers don't produce fruit. Plants do. Farmers make the conditions right: crop rotation, water, sunlight. They facilitate. They tend. They stay out of the way.

Context Farmers do the same in the new world of AMCs running on Company Brains. They monitor context quality and decay. They add context from the physical world only humans can access. All with the objective of feeding the Company Brain so the fruits of their business can grow.

It's what great executives already do with high-performing teams. As the saying goes: hire great people and get out of their way.

The people I know doing this well have ample free time. They take open-ended meetings. They collect new ideas and feed them back to their teams. They're not busy. They're farming.

The Trillion-dollar Terraform Project

The biggest challenge for Context Farmers today is the same one farmers faced in the region that would become Tel Aviv in the early 1900s: starting from nothing.

The land was a barren desert. No water, no fertile soil. If you wanted to grow anything, you'd have to figure out irrigation, fertilization, and everything else yourself.

This is exactly what context farming is like today.

To collect, monitor, and improve context for a brain, people are on their own: setting up databases, ingestion pipelines, and other systems from scratch. There are plenty of open-source tools, but the ecosystem is still early enough that they tend to fall quite short.

Back to Tel Aviv. 100 years later, it's been completely terraformed. Israelis invented drip irrigation and a handful of other technologies that turned the country so green you can see the difference between it and its neighbors from space.

The invention that changed everything wasn't a better plant; it was better infrastructure that benefited the entire farming industry.

This is the trillion-dollar opportunity of our lifetimes: terraforming the context layer by making personal and global context accessible to all.

Karpathy put it well, though I'd call it an understatement: "I think there is room here for an incredible new product instead of a hacky collection of scripts."

We need to evolve beyond MD files on our computers. Beyond company brains hosted in the cloud. To truly unleash the Intelligence Revolution, empower agents to achieve their full potential, and birth millions of Agentic Micro Companies, we need a singular global brain that serves the entire ecosystem.

This has been attempted by virtually every massive company:

• Google organized the world's websites.
• Meta organized the world's people.
• LinkedIn organized the world's professionals.
• Salesforce organized the world's deals.

But all of them fall short because they are siloed and idiosyncratic. Hand-crafted for their singular use cases.

They are not composable.

Composability, or the ability to combine independent parts into new systems, is crucial because it has been the catalyst for virtually every technological revolution in history.

Year	Revolution	The composable unlock
1455	Print	Movable type instead of hand-scribing each page
1820	Industrial	Interchangeable parts instead of custom craftwork
1991	Web	HTTP + TCP/IP instead of proprietary networks
Now	Intelligence	Composable context instead of siloed tools

The Intelligence Revolution will be catalyzed not by intelligence itself, but by composable context. A global brain powering millions of AMCs and trillions of agents.

With such a system, any information, software, or business outcome would be a prompt away. Much like in a fertile ecosystem, you just need to scatter seeds.

When it exists, context farming stops being a skill only technical founders in San Francisco can access. It becomes the default mode of operating any company.

And in this world, even Context Farming may become obsolete.

Until then, happy harvest!

---

I started my career in Computational Cognitive Science, studying early language models and how the brain processes information. Then big tech, then startups, then deep into the CRM space. A few years ago I had an epiphany: email is the most robust system of record on the planet, the right foundation for something much more ambitious. That's why I started Micro. It wasn't until that dinner in San Francisco that I realized what we'd actually been building toward.