Something subtle is happening inside corporate finance: software is moving from supporting the work to performing portions of the work.
01 / The Shift
The finance stack is becoming part of the finance team.
Andreessen Horowitz recently described an emerging CFO archetype built around the ability to design the operating system the company runs on: data, workflows, agents, skills, and controls that turn context into intelligence and decisions.
The underlying technical shift is larger than finance. Traditional enterprise software mostly organized work performed by people. The new generation can extract data, reconcile records, build scenarios, conduct follow-up, draft analysis, and execute bounded workflows.
That means software is becoming an actor inside the operating model rather than only an interface around it.
Architecture shift
The important distinction is not “AI versus no AI.” It is whether software is reading, recommending, or acting. Each step changes the authority and evidence the system must preserve.
02 / Execution Stack
From application stack to execution stack.
A simplified traditional enterprise pattern looks like:
Human → application → database
The person performs the work. Software records, organizes, or accelerates it.
An agentic finance system adds a different layer:
Human → policy → agent → tools → applications → data → action
Now the system can gather context, reason across sources, invoke tools, route work, and potentially change state.
The design problem therefore moves beyond whether an application functions correctly. The system must answer a different set of questions: what data may the agent access, which tools may it invoke, what authority does it have, what requires approval, what evidence survives, what happens when confidence is low, and how an action is reversed.
03 / Governance
Finance makes the agent-governance problem obvious.
Consider an agent preparing a journal entry. Producing the entry is only one part of the system. The enterprise also needs the source evidence, calculation logic, authorization boundary, approval chain, and audit history.
The same is true for collections, procurement, forecasting, vendor diligence, treasury, and capital planning. As a16z notes, finance needs a trusted chain from the number back to its source and clear approval rules when agents participate in the work.
This illustrates a broader BlackBoxx principle:
Capability can be black box. Authority cannot be.
A highly capable model inside an opaque operating environment is not a production-grade enterprise system. It is an uncontrolled dependency.
04 / CFO as Architect
The CFO becomes part of the system architecture.
Finance sits in a particularly interesting position because it already understands controls, materiality, authority, auditability, and economic tradeoffs. Now domain experts are gaining tools that let them participate directly in building systems.
The result is a new hybrid role: not CFO as programmer, and not CFO as CIO, but CFO as systems architect.
That means understanding enough about APIs, data models, workflows, agents, permissions, logs, exceptions, and build-versus-buy economics to define how the finance function should operate.
It also means deciding where human judgment must remain explicit. The system may prepare the forecast. Finance owns the assumptions. The system may draft the journal entry. Finance owns sign-off. The system may identify a collection exception. Finance owns escalation policy.
05 / BlackBoxx Take
The operating system becomes the product.
Karl’s analysis
The most consequential AI transformation may not be another application. It may be the gradual conversion of the enterprise itself into an orchestrated system of humans, agents, applications, models, data, and controls.
Finance is one of the first places where that architecture is becoming visible because its workflows are repetitive enough to automate, consequential enough to require controls, and measurable enough to expose whether the system actually works.
The same pattern will spread into operations, legal, HR, procurement, sales, and other functions. Each domain will increasingly include experts capable of turning operational knowledge directly into software and agents.
That is why the CFO shift matters beyond finance. It is an early example of how executive roles may change when the distance between domain judgment and executable systems collapses.
The finance function is still finance.
But increasingly, finance is software-defined.
And that is a much bigger change than giving the accounting department an AI assistant.
Source and boundary of analysis
This note draws on Andreessen Horowitz, “You Need a New CFO,” by Seema Amble, Ivan Makarov, and David Borecky, published October 8, 2026, including its discussion of finance engineering, AI-native finance tools, continuous planning, control requirements, build-versus-buy decisions, and the CFO as builder and architect. BlackBoxx’s systems interpretation is independent analysis.