Most people use AI like a messy intern..
..unreliable, prone to "creative" facts, and requiring constant babysitting.
In a professional environment, that’s a liability.
We don’t need an intern who writes emails; we need a Senior Manager who applies structured thinking, mitigates risk, and delivers work that is ready for an executive’s eyes the first time.
This newsletter is about how to bridge that gap.
The "Bad" Prompt
"Summarize these interview notes for the client."
The Result: I ran this last month on real notes. It gave me a confident paragraph, invented a $300K saving out of thin air, and completely ignored the fact that the CFO only cares about risk.
Why it fails the corporate standard:
No Audience: The model writes for "everyone," which means it writes for no one.
No Format: You get a wall of text, not a structured briefing.
No Source Control: You cannot defend the output in the review meeting.
The Big4 Prompt: The "Workpaper" Method
Every professional workpaper starts with three things: Context, Objective, and Rules.
Step 1 - The Permanent File
Upload two context files before you type a single word.
ABOUT_ME.md(Your role, professional standards, writing style)TASK_CONTEXT.md(Project background, specific audience preferences)
Step 2 - The Defensive Prompt
Copy and paste this structure:
Task: Draft a 1-page email summary of the attached interview notes.
Format:
Exactly 3 bullets.
Each bullet = Key finding | Implication | Next step.
Total length < 150 words. Plain English only.
Audience: CFO of a manufacturing firm. Time-poor, focused on risk and cash flow.
Rules:
Cite page numbers from the notes for every finding (e.g., [p.4]).
If information is missing, write [ASSUMPTION]. Do not invent data.
Ask me 2 clarifying questions about missing data before you begin the draft.
Step 3 - The Interrogation
A high-level model will now pause. It might ask: "What fiscal period are we covering?" or "Should I prioritize operational or financial risks?"
Answer the questions, then let it run.
The Output (The CFO Briefing)
Here is what this system returned for me on a 4-page note set:
Subject: Interview summary — supply chain risk
Key finding [p.2]: Vendor lead times up 22% in Q1. Implication: $400k working capital at risk. Next step: Approve 2-week safety stock trial.
Key finding [p.3]: [ASSUMPTION] on freight cost—data missing. Implication: Margin impact unclear. Next step: Request April invoices.
Key finding [p.4]: Production line 3 downtime at 8%. Implication: Delivery slippage to key client. Next step: Schedule maintenance window.
Question: Do you want this escalated to the ops committee this week?
Workpaper Tip: The Prompt Log
In my work, if it isn't documented, it didn't happen.
You may also treat your prompts like an audit trail. Use a single-line log in Excel or Notion to stay defensible:
ID | Date | Input File | Purpose | Model | Risk Level |
P-001 | 2026-05-02 | Interview_Notes.pdf | CFO Summary | Claude 4.7 Thinking | Medium |
If your manager asks, "Where did this number come from?" you have the answer in 10 seconds.
Your Turn
Reply to this email with "EMAIL" and I will send you my two starter templates:
ABOUT_ME.mdTASK_CONTEXT.md
Copy them, edit them for your specific role, and you’ll have a reusable, Big Four-grade prompt system by Monday morning.
Next Sunday, I’ll share the Prompt Log Template that top-tier teams are using to manage AI agents.
— Nam
ex-Deloitte | Big4Prompt
P.S. Writing a good prompt is like training for a triathlon: if you skip the base layer (Context), you’re going to hit the wall at Mile 20. Don't let your AI "bonk" on a deliverable.
News Roundup
The Rise of "Prompt-to-File" Standard
The "Prompt-to-File" standard is effectively ending the era of manual copy-pasting as Google and Microsoft consolidate power by integrating native document creation directly into the work surface.
Google Gemini now allows users to generate and package fully formatted files within the chat interface, while Microsoft Copilot has shifted toward "agentic" capabilities that bridge the gap between communication and production.
High-priority updates like Copilot’s "Researcher" enable one-click conversions into polished PowerPoint decks or infographics, and new local data integration allows for editing Excel workbooks without the previous cloud requirements.
Frontier Defense: The Gated Intelligence Era
The pace of LLM capability growth is forcing a shift toward "defense-contracting" style delivery for top-tier models. OpenAI has restricted access to its latest frontier cybersecurity model, GPT-5.5-Cyber, a move that mirrors the "anti-open" strategy it publicly criticized rival Anthropic for adopting with the restricted release of Claude Mythos.
This shift highlights a growing industry consensus that certain autonomous capabilities, esspecifically those able to discover zero-day vulnerabilities across major operating systems, are now considered too dangerous for general release. Access to these tools is becoming a privileged asset, gated behind "Trusted Access" programs and rigorous identity verification.
Reported by TechCrunch
The Growth Gap Revealed
A recent study by PwC reveals a stark and widening divide in the corporate world: nearly 74% of all economic value from AI is being captured by just 20% of organizations.
While most firms remain stuck in "pilot mode" focusing on incremental productivity gains, the leading 20% are using AI as a catalyst for business reinvention and identifying new revenue opportunities. These high-performers are nearly twice as likely to use AI in advanced, autonomous ways and have established robust "Responsible AI" frameworks to automate decisions safely at scale.
For professionals, the takeaway is clear: efficiency is the baseline, but true ROI is currently found in growth-oriented deployment.
Read more from PwC