A CEO's Unfiltered Takes in the Age of AI
"The org chart had eleven names on it. Eighteen months later it had one."
I spent eighteen months replacing my own corporate development team, my accounting function, and most of my own job with AI agents. This is the honest account — what worked, what it cost, and what was left when the org chart had one name on it.
Written from inside the function, not about it from a distance.
Six Saturday mornings. That's how long it took to build a process that did the work of a three-person analyst bench, for under two hundred dollars a month.
Here is the operating reality most finance leaders are quietly sitting on. The close still takes ten to twelve days, run by a team sized for a workflow that no longer exists in its original form. The corporate development function still staffs a junior analyst bench to produce screens, models, and first-draft memos — work that current tools can now do, end to end, in under an hour, at a cost that rounds to nothing against the loaded cost of the team producing it. The offshore accounting relationship that was supposed to be the cost-efficient option is now, itself, the most expensive way to do bookkeeping that current technology supports. None of this is a secret inside finance organizations. It is the thing nobody on the leadership team wants to be the one to say out loud, because saying it out loud means deciding what to do about it, and deciding what to do about it means having a conversation with real people about real jobs that most executives would rather defer.
That deferral has a cost, and the cost compounds. Every quarter a function runs on its old headcount while the capability gap widens is a quarter of margin given away to organizations that made the harder call earlier, and a quarter closer to a board member, an investor, or a competitor asking the question directly instead of waiting for leadership to raise it first.
"The gap between what AI can already do and what organizations have actually deployed isn't a technology gap. It's a decision gap. This book is what it looks like to close it — and what it costs."
I wrote the memo that would have ended four careers in one move. I almost sent it. This book is about the eighteen months between that memo and the org chart I actually run today — one name at the top, seven agent stacks underneath, and a much harder answer to the question of what was actually worth keeping.
This isn't a vendor pitch and it isn't a doom book. It's a CEO's actual ledger: the corporate development team that trained its own replacement, the executive assistant function that runs on five coordinated agents, the offshore accounting team I should have told sooner, the financial advisor relationship worth keeping and the one that wasn't, and the six weeks I spent trying to automate my own job — and why I stopped.
Every role on the cover is a real chapter in the book. Hover or tap a plate to see what happened to it — and where to read the full account.
Vera asked the question four seconds into the meeting, before I had finished my second sentence, and I have replayed those four seconds more times than I have replayed almost anything else in my career.
I had built two slides. The first showed our corporate development function's deal volume for the trailing twelve months. The second showed the same volume, produced by an AI-assisted process I had built over six weekends, at a fraction of the cost. I had practiced the transition between the two slides. I had not practiced what came after, because I had assumed, with the specific confidence of someone who has just spent six weekends being very impressed with himself, that the slides would speak for themselves.
Vera looked at the second slide for four seconds. Then she asked: "Which of these deals actually closed because of something a person did, not the model?"
I did not have an answer. I want to be precise about what that means, because it is not the same as having a bad answer. I had built two slides comparing cost and output volume, and I had not built — had not even thought to build — the slide that would have answered the only question that mattered, which was not how much cheaper the work was, but how much of the work that mattered had ever been the work I was comparing.
This book exists because of those four seconds. Everything that follows is the slow, sometimes embarrassing, occasionally costly process of building the answer I did not have in that room.
What differentiates this from the rest of the AI-and-work shelf:
No context. No setup. Just the sentences readers keep coming back to.
I had built two slides comparing cost and output volume, and I had not built the slide that would have answered the only question that mattered.
Chapter 1
The agent has nothing to lose. That sentence is the shortest, most complete explanation of why the CEO function has not been automated.
Chapter 47
Someone is going to run this analysis on your team. The only decision you have is whether that person is you.
Chapter 13
Gratitude from the trusty toward the warden is stranger and more real.
Interlude III — The Raise
The relief told me I had found the ceiling. What I had actually found was that the ceiling was higher than I thought.
Chapter 2
The twenty percent is not twenty percent of the hours. It is twenty percent of the hours and one hundred percent of the value.
Chapter 4
I had automated the homework. I had not gone anywhere near the exam.
Chapter 2
Catching it just proved the model needs exactly one human babysitter — which is why they alone were kept, which is the whole reason they can never leave.
Interlude II — The Flux Review
The most honest meeting about the firm's economics that anyone could remember having.
Chapter 9
What is left is shrinking. Not toward zero, but toward a smaller number than it was eighteen months ago — and the rate is accelerating.
Chapter 41 — What's Actually Left
Independent research on AI task coverage by occupational category shows a consistent pattern: theoretical AI capability in business, finance, legal, and administrative work already sits in the 0.8–1.0 range on a 0–1 scale. Observed, actual deployment in those same categories sits closer to 0.1–0.2. That gap exists in every finance organization currently running pre-2023 headcount against 2026-grade tooling — which, at this point, is most of them.
Three forces are closing that gap faster than most leadership teams are prepared for. Private equity operators are demanding AI-enabled finance functions as a condition of platform investment, because the cost structure of the old model no longer clears their return thresholds. The agent economy has matured past single-task copilots into orchestrated, multi-step workflows — sourcing, screening, modeling, and reporting handled end to end with a human review gate, not a human doing the work with AI assistance. And corporate finance itself is evolving: the FP&A analyst, the deal team associate, and the bookkeeper are being redefined around judgment and accountability rather than production, whether or not the org chart has caught up yet.
The organizations that move first on this aren't doing it recklessly. They're doing it the way this book documents: function by function, with a named human accountable at every review gate, and an honest account of what gets lost along the way. The organizations that wait are not avoiding the decision. They are deferring it to a board member, an investor, or a competitor who will eventually make it for them.
StackedCFO works with founders, CFOs, controllers, investors, and operators to build institutional-grade finance systems — the agent stacks this book describes, built for your organization.
Weekly insights on AI and finance, FP&A, corporate finance, technical accounting, and private equity — written the same way this book is, from inside the function.
Make sure it's you. Make sure it's now.