The rise of finance AI engineers
and the death of analysts
If 2025 was the year of agents, 2026 will be the year of making them work.
I didn’t take AI seriously until early 2025.
It would hallucinate. It couldn’t do basic math.
It didn’t understand how I worked.
All valid.
I work in finance so I don’t have room for error.
Finance professionals feel the exact same way. This is the top comment on /r/FP&A regarding AI earlier this week.
But I spent all of 2025 using AI anyways and the results shocked me.
And yeah, it sucks at basic math, but that’s not the point of the tool. That’s why it can use excel formulas and python that can be validated.
My first AHA moment was in June when I used Claude Code. The first AI tool that understood how I work (in hundreds of files scattered across my computer).
Then in November I got early access to Claude for Excel, which completely changed my relationship with financial modeling.
I’ve cut my time in Excel by more than 50% every day, which will keep dropping the more I build with these tools.
But this isn’t the case for most finance teams. They’re still stuck in the “it can’t produce accurate numbers” stage instead of playing to AI’s strengths.
Teams that figure this out will close in days, embedding finance across the business.
The bridge between "what's possible" and "how we adopt it" is still being built. And companies are starting to hire for it.
Anthropic is creating a new role to solve this: Finance AI Engineer
What is a finance AI engineer?
Finance AI Engineers (FAEs) solve finance and operations problems using an engineer’s toolkit. The problems you run into every close but never have time or support to solve is what they’re here for.
FAEs solve three categories of problems:
Time thieves — The reconciliations, the data pulls, the “why doesn’t this tie” investigations that eat analyst hours.
System gaps — Where data lives in one place but needs to be in another, and humans (or Excel) are the bridge.
Knowledge decay — The institutional logic that lives in one person’s head or a spreadsheet no one can find.
These problems aren’t new. What’s new is that AI can actually solve them when we build it right.
And “built right” means built for finance. CFOs need four things from AI automation:
Auditability: every action logged and traceable.
Governance: clear controls over what AI can and can’t do.
Explainability: answers for auditors, the board, and regulators.
Error recovery: alerts when it breaks, not at month-end.
FAEs build with these requirements baked in from day one.
Why finance teams need this role today
AI is swallowing tasks from different roles. As it does, the surface area of those roles expands while other areas collapse.
The biggest impact will hit analysts and accountants who spend 60% of their time pulling numbers, reconciling spreadsheets, and answering “can you re-cut this by region?”. These roles are going away.
What’s emerging is the analyst who prompts an agent, validates its output, and designs the workflow that replaced the manual pull.
The work isn’t disappearing but moving up a level. The analyst who used to pull data now designs the workflow that pulls it. The controller who used to review entries now validates the logic that generates them.
Same judgment, different altitude.
And someone needs to build the systems that make this shift possible.
How FAEs operate
First, they create a map. The FAE won’t be closing the books. But they’ll be in the room feeling the pain. They understand who’s involved, what’s at stake, and how it’s being done today. They get a bottoms-up view of every financial process paired with top-down knowledge of how systems operate, where data flows, and the architecture of the business.
Most teams have a monthly closing checklist. Maybe who’s responsible. Very few have a map of every financial process and how they connect.
This alone is valuable. You don’t even need special skills to do it. Just be willing to ask questions and document what you find.
But the FAE shines when they use their product development and engineering skills to build solutions to these problems; solutions that survive the realities of business and financial environments.
The operating model
The work happens at three levels:
Systems — They own the “single source of truth” by encoding business logic into AI-readable formats. Revenue recognition. IC elims. The institutional knowledge that lives in spreadsheets, email threads, and people’s heads.
Workflows — They design and manage AI agents that handle tasks currently done by analysts or outsourced teams. Reconciliation, invoice processing, vendor onboarding docs, data extraction from contracts. These run on triggers, not manual initiation: a file lands in a folder, month-end arrives, or a threshold gets breached.
Enablement — They build tools that give finance professionals superpowers. A controller asks questions in plain English and gets answers from the data warehouse. An FP&A analyst requests variance analysis and gets a first draft in seconds.
The measure of success at every level: adoption.
What FAEs are not
FAEs need to operate at the right level. Determining what they don’t do is just as important as what they do, to ensure they’re solving the right problems:
Not judgment. They encode business logic, they don’t invent it. You still need someone who knows when to accrue, who understands GAAP, and who can explain variances to the board.
Not cost cutting. Organizations that approach this as “automate the analysts away” will build brittle systems that fail at month-end.
Not IT. FAEs sit in the close, not in shared services. They feel the pain because they’re in the room.
It’s a balancing act that when executed correctly can transform a finance function, solving our biggest headaches. The problem then is finding someone fit for the role.
Progression through craft
There are two paths to becoming an FAE.
The first: engineers who learn finance. They know Python, they understand systems architecture, they can build anything. But they’ve never felt the pressure of day five of close when the consolidation still doesn’t tie. They don’t know which variances the CFO will ask about before they do. They build elegant solutions to problems they’ve only read about.
The second: finance pros who use AI to get technical. They’ve lived the pain. They know that the real problem isn’t the reconciliation but the three upstream data issues that cause it. They understand why the controller pushes back on “just automate it.” They know which process changes will survive month-end and which will get abandoned by day two.
I’m making the case for the second path.
Finance is best at solving its own problems. Always has been. ERP implementations succeed when finance owns them, not when they’re handed off to consultants who disappear after go-live.
AI will be no different.
The technical gap is smaller than it looks. Claude has collapsed the distance between “I understand the problem” and “I can build the solution.” What used to require a CS degree now requires curiosity and iteration. The finance knowledge? Debits and credits can be taught but the pain of 20 tab workbooks that never ties has to be lived.
What’s also unique about the FAE role is it offers something most finance careers don’t: progression through building, not just managing. You stay close to the work. You solve the problems that have haunted every close for years. You encode the institutional knowledge that currently lives in one person’s head or a spreadsheet no one can find.
Same judgment, different altitude. Same career, different leverage.
And because you’re sitting at the intersection of finance, operations, and AI. Three domains that rarely overlap in one person, you’re scarce. Scarcity gets compensated.
If that sounds like the career you want, start with one process. One reconciliation, one data pull, one “why doesn’t this tie” investigation. Map it. Build something small that solves it. Prove it works.
That’s how you become the person who transforms finance.
Catch ya next week! - Dave 🎣
Clear-eyed framing of how AI is reshaping finance work
You’re onto something here!!!