Netflix CPTO Elizabeth Stone says the most important skill she hires for is no longer deep specialization. It is systems thinking.
She is right. And it surfaces a question that rarely gets asked.
If seeing the system is the scarcest skill in the AI era, where is the system written down?
The Missing System of Record
Walk through any company's software stack.
- CRM holds the customers.
- ERP holds the money.
- HRIS holds the employees.
- Jira holds the work.
- The wiki holds the knowledge.
Every company has a system of record for everything except the system itself.
How decisions actually get made. Who trusts whom, and how much. Which relationships carry the business and which are quietly going cold. Where influence really flows, whatever the org chart says. That layer runs the company, and it is recorded nowhere. It lives in people's heads, and it walks out the door with every departure, every reorg, every acquisition.
For decades that was survivable. The only intelligence reading your company was human, and humans fill gaps with judgment and hallway context. A new VP spends her first ninety days rebuilding the picture: who to ask, who to convince, which approval is a formality and which one is the real decision. The company pays for that reconstruction over and over, and calls it onboarding.
Agents can't do the reconstruction at all.
What AI Sees, and What It Misses
Point AI at your documents and it understands your documents. It does not understand your organization. It cannot tell that a renewal is safe because of a ten-year relationship, or that a deal died because trust did, or that the fastest path to a decision runs through someone three boxes away from the official owner.
The most consequential layer of the enterprise is invisible to the systems we are now asking machines to reason over.
This is the argument I keep coming back to in Signal. Greatness comes from systems, not heroics (Stop Chasing Greatness. Start Building the System.). AI transformation starts with making the organization understandable to AI (AI Doesn't Replace Human Intelligence. It Runs on It.). The ontology work in that second piece covers the operational layer: objects, properties, links, actions. But there is a layer underneath the operating model that the ontology conversation usually skips. The human one. Relationships, trust, influence, and the informal decision paths that determine whether anything in the official process actually moves.
Stone is describing the org-design side of the same shift: hire the people who can see the system.
But a hire is not infrastructure. The map a systems thinker builds lives in their head, and it leaves when they do. The missing piece is a system of record for what they see.
A Graph You Can Only Build With Consent
That is the problem I am building Revolv against: relationship intelligence, a working model of the human system. Who knows whom. What the history is. Where trust is strong and where it is decaying on a schedule.
The obvious way to build that graph is the wrong way. Mine the inboxes, scrape the calendars, infer the relationships from metadata nobody agreed to share. It has been tried, and it fails for a structural reason, not a technical one: a trust graph built on surveillance destroys the thing it claims to measure. The moment people learn the system is watching them, they route around it, and the graph goes dark exactly where it matters most.
The only durable version is built from what people choose to capture. The context someone writes down after a meeting because they want to remember it. That is slower. It is also the only version that stays true.
The Next System of Record
The last generation of enterprise software recorded transactions. The next one has to record how the organization actually works.
Systems thinking is the skill. The relationship graph is the infrastructure. Companies will need both, because the thinkers keep leaving, and the machines are already here.
For companies staring at this gap now, this is the work I do hands-on through Revolv Advisory: making how your organization actually works legible, to your people and to your AI, before you ask agents to reason over it.
