The Factor's Ledger

A merchant would hand a nineteen-year-old a letter of credit, a vague set of instructions, and put him on a ship to Alexandria. Once there, the kid was expected to buy spices, extend credit to strangers, and commit massive sums of his employer’s capital in a language he barely spoke.

He was making high-stakes judgment calls thousands of miles away from anyone who could check his work.

The Venetians called him a factor. He was, in the most literal sense, an agent.

If he made a terrible trade—or just stole the money—his bosses wouldn't find out until the ship came back empty a season later.

But they sent him anyway.

They didn't do this because teenagers in the 15th century were unusually trustworthy. They did it because trade was incredibly lucrative, and the merchants of the Mediterranean had figured out the answer to a question every boss eventually faces: How do I trust what I cannot see?

Their answer wasn’t better training. It wasn't stricter instructions. It was an institution. It was double-entry bookkeeping.

When the mathematician Luca Pacioli finally wrote the system down in 1494, its brilliance wasn't just that it recorded math. It was that it engineered trust. Every transaction appeared twice, in two places, tied to each other. Money couldn't move without leaving a shadow.

This meant the books could be interrogated. An omission didn't just look like a mistake; it physically broke the ledger.

Before double-entry, delegating wealth was a bet on a person’s character. After it, delegation was a bet on a system. Even a dishonest factor operated inside a web of cross-checking records that would inevitably expose him. Historians consider this accounting machinery the quiet foundation of modern capitalism. Capital could finally travel, because evidence traveled faster than ships.

Notice the shape of what they built, because it repeats everywhere.

When human beings figure out how to delegate important work to something they can't see, they get nervous. And to calm that nerve, they build the exact same three pillars:

A record. What actually happened, preserved independently of the doer. A standard. The house rules of what was supposed to happen. A verification. A routine moment where someone compares the record to the standard and asks if they match.

Every industry with high stakes eventually reinvents this.

Aviation did it with the black box. In the 1950s, early jetliners like the de Havilland Comet were falling out of the sky and investigators had nothing to go on. An Australian scientist named David Warren suggested putting a recorder on every plane. Pilots hated it. Airlines hated the cost. But the crashes kept coming, and the requirement became universal. The black box doesn't prevent plane crashes. It just guarantees that failure is legible. And legibility is why flying became safe.

Medicine did it with the M&M conference. Since the early 1900s, surgeons have gathered in rooms to dissect their own bad outcomes line by line. No dashboards, no sampling. Just professionals grading each other's actual decisions against a standard, continuously, to turn private errors into shared knowledge.

Software engineering did it with the postmortem. When a system goes down, the timeline is reconstructed from logs, checked against policies, and verified with action items.

Records, standards, verification. When the stakes are real, civilization converges on this architecture because nothing else works.

You can see exactly what happens when one of those pillars is missing.

In August 2012, the trading firm Knight Capital pushed bad code to production. They had no reliable record of what their own systems were doing in real-time. In 45 minutes, they lost $460 million. They didn't lose money because the market turned against them; they lost it because they couldn't see what their automated agents were doing.

Or look at the British Post Office scandal. For fifteen years, the Post Office prosecuted over 900 of its own managers for theft. The evidence was simply the output of their new Horizon computer system. The system was wrong, and the humans were right. But the institution treated the machine's records as infallible. An oversight system is only as good as its independence from the thing it oversees. If the machine grades its own homework, that isn't oversight. It's just a mirror.

Which brings us to artificial intelligence in 2026.

Today, we are doing the exact same thing the Venetians did. We are handing a letter of credit to an agent and sending it out into the world. It talks to customers, amends bookings, issues refunds, and moves money. It acts at distances of abstraction instead of kilometers, makes decisions in milliseconds instead of seasons, and can err at a scale a teenage merchant couldn't dream of.

And yet, the infrastructure watching these AI agents was built for a different era. Our monitoring tools are designed to answer, “Are the servers responding?”

Yes, the servers are responding beautifully. But what did the agent actually say? What data was it looking at when it made that decision? Was that decision consistent with how we run our business?

For most companies running agents today, there is no answer. Because there is no record.

The Renaissance answer applies today because the problem hasn't changed. To make AI agents safe, you need aviation-grade legibility for every session. You need house rules translated into standards a machine can grade. And you need a closed loop that compares the two, forever.

The Venetians ran global operations with teenage agents and medieval communications, and their fraud-adjusted loss rates would embarrass most modern tech companies. They pulled it off because their oversight was structural, not heroic.

AI agents are the new factors. They just need the new ledger.