What Is a System of Intelligence? A Guide for Payment Teams

The first time someone crosses a field, there's no path. Just grass. The tenth person follows the faint line the first nine left. By the hundredth, there's a trail anyone can walk without thinking.

Most payment operations still look like that field before anyone's crossed it. Every case starts in the grass.

A system of intelligence is how the trail gets worn in.

The short answer. A system of intelligence is the layer that sits above your systems of record and turns the work done on that data into knowledge the next task can use. Systems of record store what happened. A system of intelligence learns why it happened and what to do next. Every case it handles makes the next one faster, cheaper, and more consistent. Runtime is a system of intelligence for payment teams, the governed infrastructure beneath agents that turns every investigation into compounding organizational memory.

Where the Term Comes From

Jerry Chen at Greylock popularized the idea in his 2017 essay The New Moats. He described three layers of enterprise software. Systems of engagement on top, the apps and chat where people work. Systems of record at the bottom, the CRM, ERP, and HR databases that hold the truth. And in between, systems of intelligence, which pull data across those records and get smarter as they're used.

His argument was that the middle layer would become the new moat, because the value compounds. He revisited it in 2023, when foundation models made these systems far easier to build.

What's changed since then is that agents can now do the work, not just recommend it. That turns a system of intelligence from a dashboard into something closer to an experienced teammate.

Three Layers

For a payment company, the layers look like this:

The first two layers are well covered. You have a ledger. You have workflows. What most teams don't have is the third layer written down anywhere. It lives in the heads of your best analysts, and it walks out the door when they do.

We go deeper on each pairing in system of record vs system of intelligence and system of action vs system of intelligence.

One Payment, Four Teams

Here's why this matters so much in payments.

A merchant's payout doesn't arrive. Support gets the ticket and asks payment ops. Payment ops traces it across the processor, the ledger, and the bank file, and finds it was held after a bank account change. Finance sees the same payout as a reconciliation break. If the account change looks suspicious, compliance reviews it too.

That's four teams, four investigations, and four tools, all looking at one payment. None of them can see what the others already learned.

With a system of intelligence, the first investigation leaves a trail. When support gets the next "where's my payout" ticket for that merchant, the agent already knows about the account change, the hold, and who approved the release. Finance's break already has its explanation attached. Compliance starts from the evidence, not from scratch.

Governed Memory

Memory alone isn't enough in a regulated business. An agent that remembers things nobody can trace is a liability. So the memory has to be governed:

• Permissions. Each team and agent only sees what its role allows, even inside shared memory.
• Provenance. Every learned fact links back to the case, query, and person that produced it.
• Audit. Every run is stored end to end and exportable for an auditor or your sponsor bank.
• Ownership. The memory lives in your environment and stays yours.

That's how we think about Runtime. Agents are the surface. Runtime is the governed infrastructure beneath them that turns every investigation into compounding organizational memory.

In practice, it shows up in a few ways. Agents learn from resolved cases and propose updates to their own skills. Evals check those updates against real cases before they ship. Routine work moves to smaller, cheaper models as the skills get sharper. And the agent harness keeps every step inside your cloud, with a person approving what matters.

Frequently asked questions

What is a system of intelligence?

A system of intelligence is the layer that sits above your systems of record and turns the work done on that data into knowledge the next task can use. Systems of record store what happened. A system of intelligence learns why it happened and what to do next, and gets better with every case it handles. Runtime is a system of intelligence built for payment teams.

Who coined the term system of intelligence?

Jerry Chen of Greylock popularized it in his 2017 essay The New Moats, which placed systems of intelligence between systems of engagement, like apps and chat, and systems of record, like CRM and ERP. He updated the essay in 2023 as foundation models made these systems far easier to build.

Does a system of intelligence replace our ledger or core banking system?

No. Your ledger, processor, and core banking platform stay the source of truth. A system of intelligence reads from them, connects what they don't connect on their own, and writes back only through approved actions.

How is a system of intelligence different from a data warehouse or BI tool?

A warehouse and BI tools help people look at data. A system of intelligence does the work on that data and remembers how it went: which evidence settled a dispute, which processor code meant a retry was safe, which merchant pattern preceded fraud. That memory shapes the next case directly.

Who owns the memory in Runtime's system of intelligence?

You do. It lives in your environment, under your permissions, with the source of every learned fact on record. It's never shared with other Runtime customers.

Wear the Path

You can't shortcut a trail. It gets worn in one crossing at a time.

The teams that start now will have thousands of investigations behind every agent a year from now. The teams that wait will still be walking through grass. Pick one queue and start walking.