Your finance team is talking about AI, but are they actually building with it?

SupERPower Hour is a free, live series for modern finance teams, covering prompting, vibe coding, app building, and AI agents. We build everything live using free tools. No technical background required.

See where your team stands. Join this Friday.

Register now

And they did it safely…

In February 2025, CEO (at the time) Doug McMillon told investors that AI coding tools had saved Walmart's software engineers 4 million developer hours in 2024 alone. That's equivalent to ~2,000 FTE engineers.

Then, in July, Walmart's CTO, Suresh Kumar, published a post that showed just how deep they’d got into adopting AI.

They had been building agents fast, "for every part of the business." And they'd hit a new problem: too many of them. "Multiple agents — even if each one is useful — can quickly become overwhelming and confusing."

That's management speak for two things:

So they made what Kumar called "a deliberate choice.” Consolidate everything into four super agents, one per domain, company-wide:

Each super agent has specialist sub-agents living inside it. WIBEY alone sits on top of 200 underlying agents.

While most CFOs are still working through privacy and compliance concerns, Walmart handed the keys to the organization and said, “Go build.”

But Walmart isn't dumb… they wouldn’t take undue risk, letting unproven tech run loose in the guts of their business.

So, how did they do it?

The answer is ‘Element’: Walmart's proprietary machine learning platform, built in 2021. Four years before the super agent announcement.

Element has compliance, security, and ethical safeguards built in as standard. Full observability into agent behavior: every decision path, every reasoning step, every tool the agent uses. An evaluation framework designed to ensure agents are, in the words of Walmart's EVP of Global Technology, "doing what they're doing, and nothing more."

They were more ready than you or I for AI. They simply saw it coming sooner than we did and were ready.

It’s a masterclass in using governance and guardrails to unlock innovation. It’s incredibly impressive in a business as big, complex, and mature as Walmart.

Sidenote: I would love to see the prompt for Marty, the supplier agent. My best guess: "Tell the supplier they are too f*cking expensive, and we will delist them unless they cut prices by 5%. Repeat until they start crying. Once they reach tears / mental breakdown (whichever comes sooner), shake their hand and thank them for their continued partnership. Make no mistakes."

I chose the title for this series carefully.

AI is the hottest topic in finance right now… and for good reason. If you haven't had a hands-on "holy sh*t" moment with it yet, you're sleeping at the wheel and aren’t being curious enough.

But we are at, or approaching, the peak of a hype cycle. And that frenzy has made it genuinely hard to separate what's real from what's noise.

The prompt peddlers are everywhere. The transformation consultants are circling. And serious CFOs are struggling to get clean answers to the operational questions that actually matter.

That's what this series is for.

I don’t have all the answers (don't believe anyone who says they do), but we can cut through the noise together and figure out what steps CFOs should be taking on AI right now.

If you are unfortunate enough to have a login to LinkedIn, you will know folks are losing their minds over there…

It’s easy to feel behind on AI right now.

There’s an entire industry built on making sure you feel that way.  Behind every free prompt library and AI masterclass is very smart digital marketing.

Often 100x smarter than the content itself…

That's not to say there's no value in it. But here's the truth: these tools (Claude for Excel, ChatGPT, CoWork and the rest) are spectacular precisely because they are extraordinarily easy to use. You don't need to be technical. You don't need a prompt library. Almost anyone can get useful output within minutes.

I've watched a founder with a very limited understanding of finance take a financial model spec and build it themselves in Claude. Saved themselves thousands of dollars and several weeks. It wasn’t perfect, but it was good enough for what they needed.

I’m not advocating this approach - of course - but it does illustrate a bigger point.

There is zero career alpha for a finance leader in ‘prompting a dashboard in five minutes’. That skill will be commoditized to zero. Fast.

The CFO moat isn't in knowing how to use the tools. It's in knowing what to do with them at scale, across complex workflows, using data that needs protecting.

I'm a prolific Anthropic user myself:

I'm nowhere near an expert, but that's the point. You don't need to be… it’s easy.

Most of the content you’re seeing about AI is focused on personal productivity. Faster emails, quicker models, dashboards in five minutes.

Useful? Yes. Fun? Definitely. Material for CFOs in the long-term? Not really. CFOs don’t optimize individuals. They optimize systems to unlock whole organizations. And that doesn’t come from better prompts, it comes from workflow redesign, shared systems, and institutional capability.

The CFO grade questions are more like this:

Reading this, you might be thinking, "God, who is this miserable old dinosaur? I can't wait to prompt his ass out of existence. The enterprise tiers have data privacy built in."

Do they? They say they do…

This week, Anthropic (the company behind the tools I just told you I use personally) accidentally leaked the source code behind Claude Code. Nearly 500,000 lines. Within hours, it had been mirrored and forked across GitHub tens of thousands of times. Their CCO called it a "process error" related to moving too fast. Their words, not mine.

OK this was their own data, not customer data. But that’s not the point…

If the AI lab building the tools (and promising you enterprise-grade security) can leak… your finance function, had better have thought harder about this than a LinkedIn post telling you your data is safe ‘bEcAuSe EnTeRpRiSe TiEr’.

You might also be thinking the opposite: “Yes. Yes. YES. Thank god someone said it. We should slow this all down until we understand the risk properly.”

No. No. NO. That's not what I'm saying either.

Moving slowly carries huge risks too. You risk being left for dust as your competitors use AI to become faster, cheaper, and better than you. And it doesn’t even solve the data compliance issue (more on why in a moment).

Being a tedious laggard waiting for everything to get proven out is not the answer.

In any case… there needn’t be a direct trade-off between governance/scaling concerns and progress innovation.

Walmart proved that. They’ve managed to move faster than most, and safely, in a humongous organization.

And here's a personal example on an infinitely smaller scale.

Three months ago, my business ran on QuickBooks. I was planning to vibe-code an agent on top of QuickBooks to automate some sub-routines, a sales commission calculations, some month-end reporting, basic re-forecasting, etc.

Instead, I switched to Campfire (I like to put my money where my mouth is). This week, they launched custom agents built directly into their AI-native accounting platform. Everything I'd planned to build can now be done using a native agent inside the platform. One less integration. One less maintenance headache. One less data security risk.

I won't pretend that the decision was deliberate. I didn't know the custom agents were coming. My gut just said that leading with an infrastructure decision would be smarter than spinning up custom tools on top of older tech. And now I can build more automations, faster.

This same logic applies at any scale. It’s the architecture decisions that matter most.

A series of prompts that save 30% saving your personal time is valuable. Cute, even. But it’s also kind of an advanced form of AI procrastination.

The full power of AI comes from 10x-ing the output of your whole function on 70% of the heads. To unlock that, you've got to get into the platform decisions that unleash the power of your team supercharged by AI.

Those are much bigger questions about the system of your finance function as a whole. And you, as the CFO, need to be the builder of THAT. Not some vibecoded app that might be redundant in 3 months.

That's what we’ll tackle in this series.

So, to help you make the most of this series, let’s first calibrate where you are on the AI journey and what you might be thinking about.

Today, I’m seeing five possible levels of AI maturity for CFOs and their teams:

Let’s take each in turn:

Your data is already in the wild.

If you're one of those CFOs stuck on the starting block (too busy or too worried about privacy concerns), I have bad news for you.

You already adopted AI. You just can't see it.

If you haven't made Copilot, ChatGPT, Gemini, or Claude available to your team, they are using their personal accounts to do their jobs. Even if you have told them not to, don’t fool yourself, they are.

And consumer accounts don't carry the same data protections that enterprise tiers do. Your data is in the wild west.

Simply waiting doesn't solve your risk problem. So if you think you're being careful, you aren't.

If this is where you are, the priority is simple: get off this level as fast as possible. Create an environment where your team can use AI inside a framework you control.

AI delivers productivity, but it’s trapped in individuals.

This is where most businesses are today. Some kind of enterprise-level access to one or more foundation models, available on the desktop.

For many, that means Microsoft Copilot (the natural extension of existing architecture, the easy and safe decision for IT teams). Others have been braver and implemented enterprise-level Claude, ChatGPT, or Gemini.

This is also the stage where an AI use policy tends to appear.

Depending on how expansive that policy is (and how compliant your team is), people could be using AI for very basic purposes (drafting emails, etc) all the way through to building custom agents or apps.

Businesses in this stage will have a few people miles ahead of everyone else. A huge gap between the best individual user and the average. Nobody's measuring it, and nobody's really teaching it (an AI ‘champion’ doesn’t count).

Every shortcut, every prompt, every workflow the advanced user has figured out lives entirely in their head. When they leave, they leave with them. The organization has no institutional AI skills.

The productivity gains might be real, but as long as they are trapped inside individuals and their job specs, they are fragile.

It is also hard to capture those productivity gains into delivered efficiencies, because you can’t see them. Everyone is using their newfound time to ‘be more strategic,’ whether you asked them to or not.

You’re removing friction between people, not just within roles.

This is where it starts to get interesting.

Level 3 is when AI adoption isn’t about making individuals more productive and starts making the team better overall. That's a different thing entirely.

The biggest inefficiencies in most businesses aren't in the manual tasks people do. They're in the process inefficiencies that create that work in the first place. The handoffs. The waiting. The decision sat in one VP's inbox, holding up everyone else.

At Level 3, you unlock the ability to redesign entire cross-functional workflows using AI. And attack the friction that lives between people, not just within them.

Take the budget cycle. Instead of finance sending templates to 15 department heads, getting them back half-filled, chasing for two weeks, and manually consolidating, the whole process runs through a shared AI-assisted workflow. A process that used to take six weeks of calendar time takes two. Coordination costs are reduced dramatically. Meaning the real time can be spent on the things that matter - getting those assumptions right.

Best practice spreads across the team rather than staying trapped in the best individual. Everyone is expected to embrace AI as standard. And the efficiencies are more transparent, so they are easier to harvest.

Very few finance teams are even here yet. And it doesn't happen by accident. It needs real design…

Work is executed by systems, much more than by people.

At this level, AI starts handling the real heavy lifting in the finance team.

Super-agents attacking the high-volume, load-bearing parts of the accounting and finance function. Chomping through the work that used to consume your team's time. Led and supervised by humans handling escalation and exceptions, but the AI is doing the cycle.

At this stage, you can give your agents job descriptions that are indistinguishable from human ones. Imagine a super-agent responsible for the smooth running of your procure-to-pay function. It has sub-agents receiving purchase orders, matches invoices, flags exceptions, chases approvals, and closes the cycle.

The key distinction from Level 3: Level 3 removes the process inefficiency between people. Level 4 accelerates the cycle time of the work that remains. Here you unlock something far more valuable than productivity. You unleash a new level of speed and decision velocity across the business.

But controls matter here more than at any previous level. When an agent is touching a process end-to-end, you need to know how errors get captured and which ‘human-in-the-loop’ is responsible.

Whether you build these agents in-house (as Walmart appears to be doing) or buy them embedded in your software is a key decision point in your AI strategy (we'll cover that in full in Part 3 of this series).

While I've seen some finance teams (and software) that touch this level for certain processes, I haven't seen anyone who can claim to be here across the finance function yet.

It'll come, though…

(If you are a CFO reading this and thinking, “Hey, my team is at level 4 across the board” … please reach out, I want to talk to you!)

The finance function becomes a living system.

Level 5 is like chasing a unicorn for CFOs. Today at least.

I'm including this for completeness, but without wanting to sound like an AI sensationalist. This is the utopian destination you've been told about but didn’t believe existed (like a kind of CFO El Dorado):

The AI isn't just doing the work, it's learning from it, feeding back into it, and making itself better.

And … the things that need a human are more valuable than they've ever been.

I do believe this destination is both a long way off and technically possible with today's technology. Both can be true at once.

I won't labour this level too much… it's too abstract, and we all have far too much work to do further down the ladder.

You've likely self-located on the ladder. We’ll get into what that means this month and how to continue the climb.

Here's how the rest of the series builds from where you are:

A year ago, my advice on AI for most CFOs was to get hands-on with low stakes experiments in finance, learn the power of the technology, take plenty of software demos, but hold off on any kind of mass adoption

That ship has sailed now. It was the right advice at the time, but now its time to figure out how to bring AI into your finance team and make real moves.

The question isn’t prompts or tools. It’s how you build the environment that lets this actually scale. That’s what we’ll work through in this series.

Next week… we’ll get deeper into how AI works, when it is dangerous, and how to control for those risks.

:::::::::::::::::::::::::::::::::::::::::::::::::::
:: Thank you to our sponsor ::
:: Campfire ::
:::::::::::::::::::::::::::::::::::::::::::::::::::

If you enjoyed today’s content, don’t forget to subscribe.

Disclaimer: I am not your accountant, tax advisor, lawyer, CFO, director, or friend. Well, maybe I’m your friend, but I am not any of those other things. Everything I publish represents my opinions only, not advice. Running the finances for a company is serious business, and you should take the proper advice you need.