The Tech CFO Rewires
the Finance Team
A repeat technology CFO on what actually changes when AI lands inside a finance function, and what stays exactly where it is.
- The episode’s framing is that the shift frees people to work differently rather than removing them. The same hours get redirected.
- The processes worth handing over first are the ones senior people remember doing by hand. Tie outs, filing preparation, subledger to ledger matching.
- The stack decision changed shape. Not which system of record to commit to, but which layer solves the problem and survives the audit.
- Before an agent touches the ledger: read only first, a cycle in parallel, a traceable log, a named owner, and the auditor brought in early.
- ChatFin runs agent driven execution across reconciliations, payables, receivables and the month end close, leaving final sign off with a controller.
The short answer from this episode is that the volume of work moves and the accountability does not. Caitlin Haberberger, a repeat chief financial officer in high growth enterprise technology, sat down with Ashok Manthena for a 38 minute AgentCFO episode, published in July 2026, on what changes inside a finance function and what stays exactly where it is.
What follows is the written version of that conversation: the manual work she would hand over first, how the stack question changed shape, the hiring bar, the checks a CFO runs before an agent touches the ledger, and the board room. ChatFin’s commentary sits in the marked blocks and the final section. Every quotation here is hers, word for word.
The shift frees people, it does not replace them
Caitlin’s framing is worth leading with, because it cuts against the version of this story told most often. Writing about the episode afterward, she described the technology shift in finance as being not about replacing people but about freeing them to work differently, and said finance professionals now have the chance to step into work that was previously crowded out.
Her words on it were these. “This tech shift is not about replacing people. It is about freeing them to work differently.”
That distinction matters operationally. A team that reads automation as a headcount exercise slows it down at every review gate. A team that understands the hours are redirected finds the next process worth handing over.
ChatFin holds that split in the product. The agent assembles the work, a named person approves the posting.
The manual work she can still feel in her bones
Ask a finance leader which task they would hand over first. The answer is rarely glamorous. For Caitlin, the reference point is manual SEC filing work, which she has said the technology available today would have saved her countless hours on decades ago.
That is the honest test for any finance tool. Does it remove work a senior person can still remember doing by hand at midnight. Tie outs, footnote checks, subledger to ledger matching. None of it built judgment after the first few cycles, and all of it consumed the hours where judgment gets built.
ChatFin picks up the high volume, low argument work first. Bank statement matching, supplier invoice coding, subledger tie out.
From one revolutionary system to an ecosystem
Caitlin has watched the stack question change shape entirely. Her shorthand for it is that finance has moved from NetSuite being the revolutionary choice to an entire ecosystem of modern alternatives sitting around and above the ERP.
That changes what a CFO is deciding. The old question was which system of record to commit to. The new one is which layer solves the problem and stays defensible when the auditor asks.
| Layer | What it decides | What the auditor asks |
|---|---|---|
| System of record | Where the balance lives | Who can post, under whose credentials |
| Close platforms and capture | Task sequencing, document intake | What happens before the ledger row |
| Agent platforms | The transaction and the reasoning behind it | The interface written through, and the approver |
The practical filter she and Ashok land on is whether a tool is built for finance specifically or adapted to it. That distance shows up under audit. For your own stack, AI tools for finance teams and ERP AI agents are a starting inventory.
ChatFin is additive to those layers. Where document capture alone is the bottleneck, a capture tool is cheaper and better.
What happens to the hiring bar
If the first year of an analyst’s job was built out of the tasks now being automated, the training ground has to be rebuilt somewhere else.
Caitlin’s read is not that entry level finance disappears. It is that the expectation moves earlier. New hires are asked sooner for the thing that used to come in year three: can you explain what the number means, can you defend an assumption, can you tell an operator what to do differently next month. The reviewing skill still has to be taught, but it gets taught against machine output rather than a junior person’s first draft.
Every ChatFin workflow produces a reviewable artefact rather than a finished posting, which is what makes it teachable. A new analyst reads the rows pulled and the rule applied, then decides.
What a CFO checks before trusting an agent
Trust here is a set of controls. The episode is specific about what must be true before an agent runs anything that touches the ledger.
Read only first
Prove the agent can explain a number before it is allowed to change one.
Run in parallel
Shadow the existing process for a full cycle and compare results line by line.
Trace every action
Source, rule applied, and output need to be reconstructable months later.
Name the owner
Every automated process keeps a human whose name is on the result.
Bring audit in early
A control your auditor first sees in fieldwork is a control you will be defending, not using.
None of this is unique to AI. Agents simply produce work faster than a review cadence built for humans expects, so the controls must be designed.
ChatFin logs the request, the rows read, the reasoning, the payload and the approver, and that log exports for fieldwork. Ask any vendor, ChatFin included, to show you one.

The board conversation gets better
The last thread is what changes upstairs. When the number and the explanation arrive together, the board meeting becomes a decision meeting.
Caitlin has been in enough of those rooms to be unsentimental about it. Directors do not want more slides. They want to know what changed, why, and what the team is doing about it. A finance function that answers those three on the day the month closes has more influence than one answering two weeks later. The AI agents for the CFO office shortlist is where to aim first.
ChatFin drafts variance commentary with the drivers named and every figure traced to the ledger rows behind it, so the pack and the explanation come out of one run.
What that looks like on one process, in ChatFin’s words
This section is ChatFin’s commentary rather than Caitlin’s. The episode describes the change at the level of a function. Deployments happen on one process, so here is a first one on NetSuite.
A customer sends a remittance of 482,445.00 covering forty open invoices totalling 486,220.00. Three are short paid, by 1,250.00, 380.00 and 2,145.00, and the advice carries purchase order references rather than invoice numbers.
What the agent reads
Open receivables through SuiteQL over SuiteTalk REST against the transaction and transactionline records, plus the aged debt saved search the team already uses.
What it matches
Purchase order reference to invoice, then amount, date and residual. Thirty seven clear on a straight match. Three fall out, 3,775.00 in total.
What a person does
Approves the application and dispositions the three deductions, each carrying the invoice, the customer reference and that account’s reason code history. Nothing posts without that step.
The arithmetic is the point. Thirty seven of forty lines never needed a human, and the three that did arrived with the evidence attached. That is the shape of receivables automation.
ChatFin depends on usable master data and a governed write interface. Where the customer master is duplicated or the ERP has no supported write path, ChatFin underperforms, and we say so during evaluation.
The technology available today would have saved me countless hours of manual work decades ago. The problems it solves are ones I can still feel.
Caitlin Haberberger
About the guest and the episode
The biography below is as previously published, unchanged.
Caitlin Haberberger is a repeat chief financial officer in high growth enterprise technology. She has held the CFO seat at Go1, Mezmo, Turnitin, Amplitude Analytics and Flite, served as consulting CFO at Clever, and led finance and people operations at Hearsay Social. She started her career as an auditor at Arthur Andersen and holds a BA in Economics and Urban Studies from Connecticut College. She is based in the San Francisco Bay Area.
For the controller side of the same problem, see AI for controllers and the integration coverage.
Questions this episode raises
What does the AgentCFO episode with Caitlin Haberberger cover?
It is a 38 minute conversation with Ashok Manthena, published in July 2026, on what changes when AI lands inside a finance function. It covers why the shift frees people to work differently rather than removing them, the manual work that defined a generation of finance careers, the move from one system of record to an ecosystem, the hiring bar, the checks that come before an agent touches the close, and the board conversation.
What should a finance team check before letting an AI agent touch the close?
Five things, in order. Run the agent read only until it can explain a number before it is allowed to change one. Shadow the existing process for a full cycle. Make every action traceable, so the source, the rule applied and the output can be reconstructed months later. Name a human owner for each automated process. Bring the auditor in before fieldwork. ChatFin logs the request, the rows read, the reasoning, the payload and the approver, and that log exports.
Does finance automation reduce the need for entry level hires?
It changes what they are hired to do rather than removing the role. When the assembly work shrinks, the expectation moves earlier. Explain what the number means, defend an assumption, tell an operator what to do differently. Review has to be taught deliberately, because reading an agent’s output is a different discipline from checking a colleague’s draft. ChatFin produces a reviewable artefact rather than a finished posting for that reason.
Volume moves automatically. Accountability stays human.
ChatFin runs agent driven execution across the operational core of finance, covering reconciliations, payables, receivables, reporting and the month end close. It applies your policy, executes the workflow, and leaves final sign off with a controller. Every action is written back with its trail intact, which is what makes it hold up.
Autonomous finance only works when it lands cleanly in the system of record, so ChatFin runs on the ERP you already have, including NetSuite, Sage Intacct, Dynamics 365, Acumatica, SAP Business One and JD Edwards. Check your stack against the integration coverage.
If this episode described your team, start with a working session on one process rather than a platform evaluation.
Section photograph: a person typing on a laptop beside cash and a calculator, sourced from Pexels and used under the Pexels licence.
All images are served from the Pexels content network. No third party product screenshots, vendor logos or licensed media are used on this page.
See ChatFin running on your ERP.
Tell us which system you run and which process hurts most. We will walk through it live against your data shape.
Confidential & private