Read time: 4 minutes

Leader, welcome back!

One story this week made people cheer. Another made them nervous.
Both are telling you the same thing.

Let's start with the one that went viral.

Matthew Gallagher launched a telehealth startup from his house with $20,000 and zero employees. Eighteen months later, Medvi is on track to do $1.8 billion in sales. His only hire was his younger brother.

The numbers hold up.
$401 million in revenue in year one. 16.2% net profit margin.
Hims and Hers, a public competitor with over 2,400 employees, hit 5.5%.

Here is the part the headline skips.
Gallagher did not build anything revolutionary. Medvi is a middleman for GLP-1 weight-loss drugs sitting on top of existing platforms for doctors, prescriptions, and fulfilment. He owned the front end. AI handled the rest.

He also started messy.
Hallucinating customer service bots. LegalZoom instead of a law firm.
He fixed each thing as the money came in.

The real unlock was not the tools. Every tool he used is available to anyone reading this right now.
The unlock was his willingness to build, break, and figure it out himself rather than waiting for a perfect plan.

Now the other story.

Oracle announced it is cutting between 20,000 and 30,000 employees.
The $8 to $10 billion freed up is being redirected into AI infrastructure.

This is not a startup. Oracle is a 47-year-old enterprise software company.
Their core products are ERP, financial reporting, and data infrastructure.
Their customers are finance teams.
The people affected are not just engineers. They are the people who support the workflows your function runs every day.

These two stories are not opposites.
They are the same story from different ends of the scale.

AI is compressing what it takes to deliver output.
A two-person team can now do what used to require an army.
A 47-year-old enterprise is restructuring around that reality right now.

The question your team should be sitting with is not "will this happen to us."
It is "are we the ones building, or the ones waiting."

Most finance leaders assume their team is further along with AI than they are.
In almost every team I work with, the honest answer is Level 2.

Here is the framework we use at Prime AI Solutions to assess where a finance function actually sits.

Level 1 is AI you cannot see.
Your team is already using personal ChatGPT and Claude accounts to do their jobs. Consumer accounts, no data protections, no visibility.
If you have not given them enterprise access, you have not avoided the risk.
You have just lost sight of it.

Level 2 is AI for me.
Enterprise access is live. Individuals are more productive. But every workflow, every prompt, every shortcut lives in one person's head. When they leave, it leaves with them.
This is where most finance teams sit today and most do not realise it.

Level 3 is AI for us.
This is the inflection point. AI stops making individuals faster and starts making the team better. You redesign cross-functional workflows. A budget cycle that used to take six weeks of calendar time drops to two. The gains come from attacking friction between roles, not just within them.

Level 4 is AI doing the work.
Agents handle high-volume, load-bearing processes end to end.
Think procure-to-pay with sub-agents receiving purchase orders, matching invoices, flagging exceptions, and chasing approvals. Humans handle escalation and oversight.

Level 5 is AI as the system.
Every automation feeds back into the function. Everything that can be automated has been. The humans remaining are doing the highest-value work of their careers.

The jump from Level 2 to Level 3 is where almost every team stalls. And it is the most important transition, because Level 2 gains are fragile, invisible, and impossible to scale.

Here is the 10-minute audit.
Ask three people on your team how they use AI. If every answer describes a personal workflow nobody else can replicate, you are at Level 2.

The next move is identifying one shared process where three or more people touch the same work, and redesigning it together.
That single exercise is where Level 3 actually starts.

I run an AI audit call for finance leaders. 45 minutes.
We map your function against the five levels, identify where the biggest gains are sitting, and you leave with a clear next move. Hit reply if you want one.

Microsoft Copilot Cowork is now landing for Frontier users, and this one is worth paying attention to.

It runs on Claude models rather than GPT, which matters because Claude handles long documents and complex reasoning better for finance work.

The architecture is also fundamentally different from previous Copilot: instead of one prompt, one output, Cowork spawns multiple agents simultaneously to handle full workflows. Triage your inbox, draft a response, book a follow-up, generate a summary document, all from a single instruction.

For finance teams, the most practical starting point is month-end reporting communications. Think of it as an agent that reads your actuals, drafts the variance narrative, and prepares the email to your CFO in one go. That kind of workflow compression is where the real time saving sits.

The caveats are real though:

Right now everything outputs into a generic OneDrive folder with limited scoping controls. For any organisation where SharePoint governs document management, that is a friction point.

Governance over which data sources agents can access is also not fully solved yet. UK and EU tenants need to manually enable Anthropic models in their M365 admin centre first as they are off by default. Treat this as Frontier, not production.
Map your workflows now so you are ready when it reaches general availability.

Claude is also now available as an add-in inside PowerPoint, accessible through the Microsoft Marketplace.

For finance teams, the practical use case is board pack preparation.
You paste in your key numbers and commentary, describe the audience and structure you need, and Claude builds or reformats the slides in your existing template. No more spending two hours on layout after the analysis is done.
The analysis was always the hard part. The formatting was just the tax you paid for not having a designer on your team.

To set it up: install the Claude by Anthropic add-in from the Microsoft Marketplace, open PowerPoint, find Claude in the add-ins panel, and it opens as a sidebar.
From there you can create from scratch, build on-brand slides from an existing template, or edit what you already have using plain language instructions.
Available on Pro, Max, Team, and Enterprise plans.