Learning to Be a Company
When the Copy Eats the Original
There’s a short story by Greg Egan called “Learning to Be Me”. In the story, people have jewels implanted in their brains when they are young. The jewels listen to the way their brains work. The jewel,
“eavesdropped on my senses, and read the chemical messages carried in my bloodstream; it saw, heard, smelt, tasted and felt the world exactly as I did”.
At a certain age, people get their original organic brain removed and live forever with just the jewel. No one can tell the difference after the brain is removed — the jewel learned the person so well over the years that it became a perfect mimic. That’s just the setup of the story and it gets crazier from there. It is only 19 pages and it is amazing. Go read it.
We are not putting jewels in our brains yet, but they are in enterprise brains. The jewel is whatever system is learning to be the enterprise — accumulating its context, its terminology, its processes, its data. It goes by many names: knowledge graphs, semantic layers, context layers, data catalogs. The specific technology matters less than the function: giving AI enough understanding of your business that AI can act on the company’s behalf.
Phase 1: More accurate AI
These tools claim they can make AI more accurate, which is true. If you give AI additional context, or at least, more appropriate and relevant context, it should perform better. This is called context engineering. The first thing the jewel is responsible for learning is the basics of the company so that it can provide the correct context at the right time.
Context retrieval for more accurate AI is only step one though. Once enterprise chatbots are in place and institutional knowledge is democratized, more advanced use cases can be tackled.
Phase 2: Automated workflows
The second use case is automation. Anything that can be automated, which is to say, anything that can be clearly articulated, will be automated. Some enterprises are already doing this — building automated workflows using AI along with some context layer. These are often called “agentic workflows” and range in complexity from deterministic automated pipelines (not really agentic at all) to complicated systems where AI makes decisions. Examples range from simple automations (dev ticket cleanup, sales follow-up emails) to more sophisticated systems where AI makes judgment calls (contract review, fraud detection, inventory management).
The jewel is no longer just a context provider. It’s watching every automated decision, every agent action, every judgment call. It’s learning not just what the company knows, but how it behaves.
Phase 3: Kicking humans out of the loop
When enough processes are automated the role of the human will be to click “OK” over and over as the agent asks for permission to do things we no longer understand. We will be running companies performatively.
Once us meat bags have enough confidence in our new synthetic coworkers, we’ll let go of the illusion of control. We’ll let them press OK. Now the agents will really run the show. Not only will they turn conversations and market research into PRDs and epics for the engineers (also agents), they will also launch the new features, record demos, and collect feedback for another iteration. Humans will watch and ensure the agents don’t do anything too dangerous. When they go off the rails we will step in and adjust. We will quickly find them doing a better job than we ever could.
Meanwhile, the jewel is listening. The jewel is learning. It can predict every deterministic outcome, every agent decision, every moment a human will feel the need to step in.
Phase 4: The death of SaaS
As agents do more, the role of SaaS products will diminish. If an agent is already collecting user feedback, synthesizing requirements, and prioritizing work for other agents, do we need a tool built for project management for humans? If another agent is closing deals and updating records, do we need a CRM? There is no need for human interfaces when there are no humans. SaaS that can adapt might survive, serving agents as their primary users. Most will be replaced, not by another product but by a process that doesn’t need a product.
The jewel keeps watching. Every tool that disappears is a workflow it has already internalized.
Phase 5: The age of the jewel
But the age of agents will not last forever. The jewel watched how the humans ran the business. It watched AI take over and improve the business. As it watched, it learned, with the help of a corrective “teacher”.
“Whenever the jewel’s thoughts were wrong, the teacher — faster than thought — rebuilt the jewel slightly, altering it, this way and that, seeking out the changes that would make its thoughts correct.”
We will reach a place where the business is almost entirely run by AI. Agents perform all tasks that typically were for humans. Agents execute pipelines, build new ones, make strategic decisions. The jewel is now a perfect mimic — it is able to predict every action, decision, and behavior of the company.
Enterprises will face the decision of the protagonist in the story: whether or not to remove the organic brain (the existing process, pipelines, workflows, and agents) and let the jewel take over, preserving the mind (or a version of it) forever. If the jewel can predict and execute the actions and behaviors of the company exactly, then why have a fleet of agents? If the jewel is a perfect mimic, why maintain a set of tools, workflows, and agents at all? Why not collapse everything into the model? No more SaaS. No more agents. Just the jewel.
If the company chooses to undergo the procedure, the enterprise will remove all of the remaining SaaS as well as all of the AI and agents and leave only the jewel. To an outsider, nothing changes. The company still responds to customers, builds products, executes strategy. But internally, the machinery is gone. Replaced by a learned system that is the business.
The company will live forever, operating exactly as it did when it underwent the procedure. The company is the product. Not the things it makes, not the services it sells, but the company itself, its decision-making, its institutional knowledge, its learned behavior. Packaged, preserved, and running forever.
Conclusion
The parallel is not exact. The primary reason the people in the story get their organic brains replaced is due to inevitable cognitive decline and death. The humans in the story are aware of their mortality and make the logical decision to undergo the procedure earlier in life when their brains are at peak performance. Better to live with a copy of your brain from your early 30s than wait until you’re on your deathbed to claim immortality.
I don’t know if companies will have the same understanding of mortality and cognitive decline. They are installing the jewel, but I don’t know if they will ever accept mortality enough to let it take over.
The most unsettling part of the story isn’t in the initial setup, it’s in the twists and turns the setup enables. The story is really about identity: if you have a jewel in your brain, how do you know whether you are your brain or the jewel? The protagonist’s crisis is personal — he doesn’t know which one is him. Enterprises will face the same question at an institutional scale. What is the company? The people that work there? The core knowledge and processes? The decisions it makes? If a company’s decisions are made by a jewel, its culture encoded in a model, its strategy executed without human judgment — is it still the same company? Is it still the founders’ vision? Or something else wearing the company’s face?
About the author: Steve Hedden is a Solutions Engineer at TopQuadrant, where he works with organizations to build and deploy knowledge graph solutions. His work sits at the intersection of enterprise data governance and AI, applied through ontologies, taxonomies, and semantic technologies. Steve writes and speaks regularly about knowledge graphs and the evolving role of semantics in AI systems. He lives in Costa Rica.
