This is part 1 of Finance AI Field Notes, a series grounded in a July 2026 scan of ~105 live job postings for people building AI inside finance functions, plus verified market research.
Ten years ago, a "Finance Systems Manager" administered the ERP. They managed user access, configured approval matrices, ran the NetSuite upgrade, and handled the integration vendor. The job description asked for "hands-on ERP experience" and "strong Excel."
I just read about a hundred job postings for that same role family, posted in the last month. The job has changed profession.
Faire is hiring a Head of Finance Systems & Data at $236–324k whose mandate is to "drive AI enablement and modernization across the finance org — close automation to reporting to operational workflows." Scale AI's Head of Finance Systems & Automation is asked to "build and manage internal agents that automate forecasting inputs, variance analysis, close task management, and reconciliation" — and the manager under them must design "reliable, production-ready agentic workflows (validation, retries, guardrails)." Pigment's Finance Systems Engineer in Paris will "deploy workflows via n8n or APIs to automate month-end closing tasks, anomaly detection, and real-time cash forecasting."
Validation. Retries. Guardrails. Orchestration. This is software engineering vocabulary, sitting in finance-department job descriptions, hired by CFOs.
What actually changed
Three shifts, visible directly in the postings:
1. The deliverable changed from configuration to construction. The old role configured what a vendor shipped. The new role builds things that didn't exist: agents that reconcile, close checklists that execute themselves, exception queues that triage their own contents. Rohlik Group's new Head of Finance Automation — the title literally says "AI Agentification" — inherits agents already in production that "chase overdue payments, process invoices, allocate cash" and is expected to be "close enough to the work to build an agent or connect two systems through an API."
2. The reporting line stayed in finance. These are not IT roles. Faire's role reports to the CFO. Rohlik's reports to the Group CFO. Aon put a Director-level "AI Lead" inside Controllership scope, responsible for close and operations. Finance is not waiting for the CIO's roadmap; it is hiring its own builders.
3. The control expectations came along. Because the builders sit in finance, the job descriptions carry finance's obligations. Kraken's posting requires "experience deploying automation in a SOX-regulated environment, with demonstrated understanding of which controls must be preserved." Aon's asks the hire to make agents "audit-defensible and explainable" with "logging, execution traceability, and override controls." Nobody writes that in a data-science JD. It's a finance JD wearing an engineering toolbelt.
The skills gap this creates
Here's the uncomfortable part. The people who held the old role — excellent ERP administrators, integration managers, process owners — are being asked to become something adjacent but different. And the people who have the new skills — engineers who can ship an LLM workflow with retries and guardrails — mostly don't know what a three-way match is, why the intercompany elimination matters, or what an auditor will ask in March.
The postings show companies resolving this tension in different ways. Some hire the engineer and hope finance knowledge transfers (Shopfully's AI Architect, Finance requires "10+ years in AI/ML" and only "solid working knowledge" of FP&A). Some promote the accountant and hope the engineering transfers (Wasabi asks a Senior Accountant to connect "Claude to live data through MCPs and APIs"). Both bets can work. Both fail more often than anyone admits, because the role is genuinely two professions in one headcount.
If you're hiring for this role
Write the job description around outcomes with control constraints, not around tools: "cut close by N days without losing a single piece of audit evidence" beats a list of frameworks. Decide explicitly which profession you're hiring and which one you'll scaffold — with training, with tooling, with a platform that carries the accounting semantics so the hire doesn't have to rebuild double-entry from scratch. And accept that if the whole capability lives in one head, you've created a key-person risk your auditor will eventually ask about.
If you're in this role
You're early to something real. The postings say the market now pays $150–320k for people who can stand between the ledger and the language model and be trusted by both sides. The scarce skill isn't prompting — it's knowing what must not break: the controls, the evidence trail, the reconciliation to the penny. Learn the engineering. Keep the accounting judgment. That combination is the job now.
I'm collecting and reading these postings as research for Artifi, where we build the governed layer this new role builds on. Next in the series: how finance leaders actually plan to adopt AI — per the survey data everyone quotes and the hiring data almost nobody looks at.