Bessemer's research team has been doing the clearest thinking in enterprise software this decade. Their taxonomy — systems of record for the database era, systems of action for the AI-native wave replacing it — captured something real. For twenty years, anyone who walked into a VC office and said they were going to replace SAP got laughed out of the building. Bessemer correctly diagnosed why that's changing: AI degraded every incumbent moat at once.
We agree with the diagnosis. We disagree about where it ends.
Systems of record stored things. They were databases wrapped in workflow, purchased at enterprise scale because digital filing beat paper filing. Oracle, SAP, Salesforce, Workday — all variations of the same idea: give the company a place to keep its data, with structured ways to update and retrieve it. The value was custody of information. The cost was everything required to operate around that custody: the analysts translating business needs into schemas, the administrators managing access, the integrators wiring systems together, the reports people ran because the database couldn't tell you what mattered.
Systems of action are the AI-native response. They don't just store — they act. They auto-populate the CRM from your meetings. They generate the ERP configuration from natural language. They advance workflows without waiting for a human to click. Doss, Day.ai, Everest — each building a category-specific version of the same idea: AI doing work that previously required human operators of software.
This is a real improvement. Systems of record asked humans to operate software to get to outcomes. Systems of action ask AI to operate software to get to outcomes. Less human time, less error, faster results.
It's also still software.
Look at what's underneath a system of action. There's a database. There's a schema. There's workflow logic written in code. There's a module that needs to be configured for each customer. There's an implementation team. There's training. The AI on top is impressive. The thing underneath it is the same thing that's been underneath enterprise software for forty years — an artifact built by engineers, operated by people, updated through change control.
The AI makes the operation of the artifact easier. It doesn't remove the artifact.
The third era — the one nobody in the VC taxonomy has named yet — doesn't have the artifact.
A system of intelligence doesn't store the way databases store. It holds business knowledge as readable sentences, connected in a graph that intelligence can reason over. It doesn't execute the way programs execute. It reads the sentences relevant to the current situation and decides what to do, the same way a good manager would. It doesn't present the way applications present. It generates what each worker needs, for the activity they're doing, from the knowledge the system has about them right now.
There is no module to configure. There is no schema to migrate. There is no code to deploy. A CEO can read every rule the business runs on because every rule is a sentence. Change a sentence, the business behaves differently tomorrow morning.
This isn't a better system of action. It's a different primitive. The way a system of action is a different primitive from a system of record — not an improvement of it, a replacement for it.
The reason this distinction matters isn't semantic. It matters because you can't build a system of intelligence by improving a system of action. The core architecture is different. Systems of action treat the software as the product and AI as the interface. Systems of intelligence treat intelligence as the product and the minimal infrastructure underneath as plumbing. Starting from a system of action and trying to get to a system of intelligence means deleting most of what you built. The incumbents won't. The funded AI-native startups can't, because they spent their Series A building the thing they'd have to delete.
The window for building the system of intelligence is the window for building it from scratch, by people who were never going to build a system of action in the first place.
That's what AI-BOS is. The business is described in plain language. The intelligence runs it. The software layer, in the sense we've known it for forty years, doesn't exist.
Systems of record were the filing cabinet era. Systems of action are the assistant-at-the-filing-cabinet era. Systems of intelligence are the era after you notice there was never a reason to keep the filing cabinet.