My Cleaning Lady Killed My AI

She Closed the Laptop and Three Hours of Work Disappeared

Not metaphorically. My cleaning lady closed the lid on the second MacBook running my AI automation, and the agent died mid-task. Three hours of processing. Gone. And because it was running locally on my machine, I couldn't recover it until I got home.

I laughed about it for a week. Then I realized it was the most honest thing that's happened in AI this year.

Here's why. Every conference talk, every LinkedIn post, every breathless prediction about AI in finance operates at 30,000 feet. The promise. The potential. The case studies from companies with eight-figure technology budgets. But nobody talks about the sticky note you have to put on a laptop so your cleaning lady doesn't kill your agent. Nobody talks about what "80% reliable" actually costs you in a function where the other 20% is the whole job.

I'm going to.

What My Week Actually Looks Like

I'm roughly 80% vibe working with Claude Code. That phrase needs a definition. Vibe working means I talk, it executes. I describe the output I want, give it context, and it produces work. Forecasts, data analysis, decks, legal documents, policies, marketing workflows, HR documentation. Not perfectly. Not every time. But enough that my operational rhythm has fundamentally changed.

As I write this, I have multiple Claude Code sessions running in parallel. One is building a presentation. Another is updating a project plan. A third is pulling data for a dashboard.

That's not a demo. That's Tuesday.

Last Friday I built a spreadsheet with our full cash runway model. Earlier that week I asked Claude to prepare a project plan for a critical initiative, and it loaded it directly to Notion with tasks assigned. After we reviewed it as a team, it updated all the tasks and timelines. In parallel I was spinning up dashboards to track the performance of two separate operations.

That's a week of work. Done by Friday. And I still made my kid's volleyball game.

The people who dismiss this as hype haven't actually tried to do real work with it. And the people who think it's solved haven't hit 20% yet.

The 80% Problem Is a Real Problem

I've been using a personal assistant tool that has access to my emails, messages, and meeting transcripts. It gives me a daily digest that cuts through the noise in a genuinely useful way.

But it works about 80% of the time.

In finance, that sentence should scare you. An 80% reliable forecast isn't a forecast. It's a liability. A reconciliation that misses one in five entries isn't done, it's deferred. When I'm reviewing a board package and something looks off, I can't note "probably fine, the AI got most of it." That's not how fiduciaries work.

So the 80% problem is real. Accept it before you build your workflows around it.

But here's the other side of that: 80% is still a massive leap when your baseline was doing everything manually. If you were producing a monthly cash flow model in three days, and now you're producing it in four hours with AI handling the first draft and the scenario variants, you're still winning on net even accounting for the review overhead. The math works. It just requires you to stop expecting perfection and start designing for verification.

Both things are true. The tool is imperfect, and it still changed how I work. Don't collapse the tension between those two facts. They're both load-bearing.

The Gap Between Me and My Team

Here's where I have to be honest in a way that doesn't make me look great.

My finance team still lives in Excel. Power BI is their ceiling. They do solid, reliable work. The reconciliations are right. The closes happen on time. They know the business.

But the gap between what I'm doing with AI and what they're doing without it isn't a training problem. It's an infrastructure problem, a trust problem, and a culture problem layered on top of each other.

Most finance teams don't have the tooling I'm running. I've invested in it. I've configured it. I've broken it and fixed it. That's a significant time sink before you see any return.

Finance people also have a healthy skepticism about tools that produce outputs they can't fully trace. That skepticism isn't wrong. It's the same reason we audit. The problem is that skepticism without experimentation becomes calcification.

And the culture piece is harder than either of those. The people who adopt AI soonest are the ones who are already overloaded and already looking for a way out. They don't need convincing. They need permission. The people who won't adopt it are the ones who've built their value on doing the hard work the slow way. Asking them to automate is asking them to obsolete their own identity. That's not a training problem. That's a management conversation.

Closing this gap is a multiple-year shift, minimum. I don't say that to be discouraging. I say it so you stop planning for it to happen after a one-day workshop.

Both Things Are True at the Same Time

Let me put the full picture on the table, because I think the debate about AI in finance suffers from people picking a side and staying there.

I built a forecast in Claude that would have taken two days. My team still reconciles in spreadsheets. I am 80% vibe working. Most finance teams haven't started. The shift is real and it hasn't happened yet.

Both things are true simultaneously. The transformation is underway for the people who started early. The transformation hasn't landed for most organizations. You can believe both without contradiction.

What I find most useful about holding this dual view: it tells you exactly what matters right now. You don't need to wait for enterprise-wide AI adoption to start changing your personal output. You also can't declare victory on AI transformation because you started using it for your own work. Those are different conversations.

The former is about your competitiveness as a finance leader. The latter is about your organization's capacity. Both matter. They're just not the same problem.

So Who Actually Owns This

Every AI conversation I have with finance leaders ends the same way. Someone eventually asks: "Who's supposed to own AI at our company?" And the room goes quiet.

IBM found that 26% of organizations now have a Chief AI Officer, up from 11% a year ago. Companies with a CAIO report 10% higher ROI on their AI investments. That gap will only widen as the tools compound.

The instinct is to put it in Technology. That's where you put things that require technical infrastructure and ongoing maintenance. It makes sense on paper.

But here's the problem. At GAIN, Technology reports to me. When we were interviewing for a critical technology role, I talked to dozens of candidates. Most of them hadn't actually built anything with AI in their own workflows. They could describe the landscape. They could tell you the major vendors. But they hadn't vibe worked a single thing.

The best candidate for owning AI at most companies isn't a technologist. It's someone who has lived in both worlds. Someone who manages budgets, teams, technology, and operations simultaneously, and has already started doing the work.

That profile looks a lot like a CFO. It also looks like a COO. These are people who are already translating between what's technically possible and what's operationally practical. They already sit at the intersection of data, process, and organizational behavior. They're already accountable to outcomes, not just implementations.

I'm not saying every CFO should become a Chief AI Officer. I'm saying the person who earns that seat will be someone with financial and operational credibility, not just technical fluency. The language of AI governance is the language of risk, ROI, and organizational change. Finance people speak that language natively.

The Real Risk Nobody Talks About

The conversation about AI usually frames the threat as "AI replacing your job." That's not what keeps me up at night.

What keeps me up is this: the people who figured out how to work with AI are already producing at a fundamentally different rate. Not 10% better. Multiples better. They're building in hours what used to take days. They're running parallel workstreams that would have required a whole team. They're iterating on analysis in real time instead of waiting for a model refresh cycle.

The gap between a finance leader who has integrated AI into their actual workflow and one who is still waiting for the ROI study is growing every month. Not because the waiters are bad at their jobs. Because the adopters are compounding.

That's the real threat. Not replacement. Getting lapped.

And the lapping is quiet. It doesn't announce itself. You won't feel it until someone you manage presents something in an hour that used to take a team two days. Or until a peer shows up to a board meeting with scenario analysis that would have taken your team a week to build. By then the gap is already wide.

The entry point isn't complicated. Pick one workflow that happens every month and takes too long. Build an AI-assisted version of it. Verify the output yourself the first few times. Refine the process. Then pick the next one.

You don't need a strategy before you start. You need a result before you can build a strategy.

Start Before the Organization Is Ready

Transformation timelines are always longer than you want them to be and shorter than you plan for.

Your organization's AI readiness is a constraint you work within. Your personal readiness is a constraint you choose. Stop confusing them.

I'm not waiting for my finance team to adopt AI before I change how I work. I'm working differently now, which changes what the team gets from me, which changes what they're expected to produce, which eventually changes how they work. That's how culture shifts. From the front, not from a memo.

The gap between AI excitement and AI readiness is where careers get defined. The people who close that gap for themselves, before it's a company initiative, before it's a performance review criterion, before it's obvious, those are the finance leaders who will look prescient in three years.

Everyone else will be in a workshop learning what those people already do before breakfast.

P.S. The MacBook now has a "DO NOT CLOSE" sticky note on it. In Spanish and English. My cleaning lady found it hilarious. The agent is still running. For now.