I Built an AI Board Member in Cursor. Here's How.

Every month, I send a board update to my investors. Revenue, margins, burn, churn, what shipped, what didn't. I grade myself. I try to be honest about what's working and what isn't.

I wanted something that would give me honest feedback before I send it out. No agenda. No politics. Just: here's what's wrong, here's what's missing, here's what your board is going to ask you.

That's hard to get from people. Your investors are busy. When they do respond, it's after you've already sent the update. And most people soften their feedback, even when you'd rather they didn't.

So I built an AI board member.

How I set it up in Cursor

The feature that makes this work is Cursor Rules.

Rules are markdown files in .cursor/rules/ that get loaded into every AI conversation in the workspace. They're a persistent briefing packet. The AI reads them every time, so it always has context without you repeating yourself.

I have three rule files:

Agent (agent.mdc): The router. Defines the workflow and ties the other rules together. When I hand it a PDF, it knows what to do.

Company context (company-context.mdc): Everything a board member would need to know about my business. This is why you never have to answer the same question twice.

Review format (review-format.mdc): The output structure and tone. Grading rubric, summary card layout, writing style. Direct, no cheerleading, look for gaps between narrative and numbers.

The workspace is simple:

• updates/ is where processed board updates live. I give Cursor the PDF of my board deck and have it extract the content into markdown and save it here.
• reviews/ is where the AI output goes. Each review produces a detailed markdown analysis and a visual summary card with color-coded grades. I actually include the summary card in the final deck I share with the board.
• board-history.md tracks metrics across every review so the AI can spot trends over time.

No plugins, no custom tooling. A folder structure and three rules files.

Why it works

It responds instantly and doesn't sugarcoat. I don't have to brief it. I don't have to wait for it. I drop in the update and get feedback I can act on.

It catches what you stop seeing. Grammatical errors, unclear phrasing, the missing parenthesis on page 5, the repeated table on pages 16 through 18. After your tenth revision, you're blind to all of it. The AI isn't.

It asks the questions you're avoiding. "When does new business outpace churn?" is easier to read in a markdown file than in a reply from your lead investor. You can figure out your answer before anyone's watching.

It's easier to take. A C- from a markdown file doesn't hit the same way as a C- from a board member. There's no tone, no subtext. I just decide if it's right, and if it is, I figure out how to address it in the update.

It never loses context. It knows my financials, my history, my churn patterns. It won't ask me to re-explain what happened last quarter. And it won't ask the same question twice. Unless the answer still matters, in which case it should.

It makes the update better. By the time I send it out, I've already stress-tested the weak spots. I've either fixed them or I have an answer ready. The board's questions don't catch me off guard.

Try it yourself

Here's what you need to do. Just ask your AI code editor to read the below to get going.

Project layout

Board Review/
  .cursor/rules/
    agent.mdc
    company-context.mdc
    review-format.mdc
  updates/
    2026-02-february.md
  reviews/
  board-history.md

The rules files

You need three rules files in .cursor/rules/. I've put all three in a GitHub repo you can grab:

• agent.mdc — The router. Maps out the folder structure, defines the 6-step workflow (extract PDF, read history, review, generate summary card, update metrics), and sets context rules: always compare to prior periods, flag reversals, call out missing sections.
• company-context.mdc — A template you fill in with your company details, key metrics, and the review structure you want (errors and issues, performance feedback, tone). This is the only file that requires customization.
• review-format.mdc — Controls the visual summary card output. Defines the grading rubric (A through D, color-coded), three categories to grade on (operational execution, revenue performance, strategic progress), writing style rules, and examples of good summary lines. Use it as-is.

Give it a try

Open Cursor and tell the agent:

Read this article [link to this post] and set up the same workflow for my company. Here's my board deck. [attach PDF]

It'll create the rules files, process your deck, and give you your first review.

If you have past board updates, load them all in. The more history the AI has, the better the reviews get. It'll use them as context for spotting trends, tracking open issues, and asking better questions.