Write your finance context file with me (Claude skill inside)

Here's how & my file. You'll get the full 16-layer map of what to write down, the trick that stops AI inventing numbers, and the Claude file I use to write my context.

Hey Finance Engineers,

Bad news first. The smartest AI on earth still can’t read your mind, and right now the mind is, unfortunately, where most of a business lives.

Good news. I built a skill that writes that context with you. It interviews you in chat, walks you through 16 layers of your business, and hands you the finished .md file at the end. It’s the same one our finance team of 20 runs on, and it’s at the bottom of this issue.

So before you implement AI, before you scale it, and well before anyone starts a token leaderboard, you write the context file and set up the data sources. This is step 0, before you do anything else.

This isn’t just my take from inside finance. The whole AI field already moved here: in 2025 they swapped “prompt engineering” for “context engineering,” with Andrej Karpathy (co-founder of OpenAI, former head of AI at Tesla, and now a part of Anthropic) and Shopify’s CEO leading the rename.

Side effect: building this context makes the business less fragile. Right now too much of how your company runs lives in one person’s head, and the day they leave, it leaves with them. Writing the context is how you get it out of heads and into text.

FUEL | AI-native corporate finance ➚

Two rules before you write a word

It’s easy to get a context file wrong, and a bad one makes everything worse, not better. Two rules I live by.

Rule 1: context is never one-time. A business changes all the time. It’s a live organism. So you work the context all the time too. Good practice: check it at least monthly and update.

Rule 2: don’t write too much. It should be clear and precise. If something is half-formed in your head, it’ll come out exactly that half-formed for the AI.

At Fuel (my company) we thought building the agents would be the biggest challenge, and that once we had them, they’d cut 80% of the manual work.

Turns out building them wasn’t the hard part. Supporting them was. Once they exist, you have to keep feeding them the right context, the new finance rules, the new units, the new business streams, and that part never really stops.

And what surprised me most is that this context turned into a finance library the rest of the team uses. We have one at Fuel now that all 20 of our finance analysts and managers use, every day.

That’s where we’re going today. The full 16-layer map of what to write down, the trick that stops AI inventing numbers, and the builder file to do it all with. But first, the obvious question: what is context, really?

So what even is “context”

Imagine a new FP&A manager joins your team. On day one, you or your manager would sit them down and walk them through how everything actually works. If you’re lucky and someone does that for you. And if not, if that manager is you, figuring it out alone, then even better that you’re reading this.

Context is exactly that onboarding, written down. Everything you’d have to explain to a new hire on their first day.

• The business itself: revenue streams, history, types of expenses, business units

• Your key KPIs and goals

• The market you’re in

• Your chart of accounts

• How the team is set up

• And all those financial rules that live in your team’s heads and have never been written down anywhere

That last one is where most of the value hides. The undocumented rules.

If you’re a finance person, you know the case. There’s an xls someone manages, full of formulas and rules, and only the owner knows what the hell is going on in there. None of it is written down anywhere. It just lives in their head, and the day they leave, it leaves with them.

Which, when you say it out loud, is a genuinely insane way to run a company.

It reminds me of that bit in Severance (the Apple TV original), the office self who knows everything about the job and remembers none of it the second they walk out the door. Most finance teams are running a version of that.

Where the context lives

Before you write anything, know where to look. In my experience context sits in four places, and most of it is undocumented.

1. What’s already written. Existing docs, decks, prior notes, team onboarding.

2. Conversations. Call recordings and meeting transcripts. This is where the strategy and the why live.

3. Day-to-day messages. Slack and email threads, where decisions actually get made.

4. Your systems. The accounting tool, billing, payroll, the chart of accounts itself.

The written 10% is the easy part. The other 90% is in conversations, chat threads, and people’s heads.

The 16 layers of context

Here’s the structure I’d use. You don’t need all of them on day one, but this is the full map.

Foundation, who you are

1. General company context. Legal name, country, entity type, industry, business model, stage, currency, fiscal year.

2. Key facts and decisions. Strategic choices already made, known anomalies in the numbers, open decisions, and what’s coming in the next 3 to 6 months (a fundraise, a pricing change, a new product).

3. Glossary. Your terms, your metric definitions, your naming conventions. What you call things, and what you never call them.

Rules, how you operate

4. Business rules and policies. Who approves what spend, payment terms both ways, expense thresholds, capitalisation policy.

Continue reading

The good part is below: the full 16-layer map and the actual file that builds your context for you (it interviews you in chat, our whole team runs on it).

→ If you’re a vendor and you genuinely have something to offer this audience, email me at team@thefinanceengineer.io, or DM me on LinkedIn.