Back to Guide
Chapter 6 | Part 2: Building

CEO Use Cases

Real use cases from real executives. Six categories with briefs ready to paste.

9 min read

The best use case is your specific problem. These are patterns to recognize your problem in.

Who Is Already Building

Before we get to the patterns: the people who prove this works aren't hobbyists.

Sundar Pichai (CEO, Alphabet/Google) uses Replit to build personal dashboards — custom web pages that pull his preferred information into a single view. His take: "It's exciting to see how casually you can do it now." This is the CEO of a company where AI now writes more than 30% of all code — and he still builds his own personal tools himself.

Sebastian Siemiatkowski (CEO, Klarna) has no technical background. He uses Cursor to validate product ideas before involving his engineering team — building prototypes in approximately 20 minutes that previously took weeks of coordination. His description: "obsessed." The point isn't the 20 minutes. The point is he goes into meetings with his team already knowing what he wants, because he built it.

Obie Fernandez (CTO, author of The Rails Way) built what he calls a Personal CTO Operating System: custom commands for meeting transcripts, hiring pipeline management, and engineer performance tracking — all personal tools that serve his specific workflow.

These aren't demonstrations. These are working habits.

Collins Dictionary named "vibe coding" — the practice of building software by describing it in plain language — its Word of the Year for 2025.

The Six CEO Categories

1. Intelligence Aggregation

What it is: Pulling information from multiple sources into one readable view, on your schedule.

What Pichai built: A personal dashboard aggregating news and data sources into a single custom page. He didn't need a developer for this. He didn't need to learn to code. He described what he wanted and built it over a weekend.

What you might build:

  • Weekly business summary combining sales data, support tickets, and team updates into one email
  • Competitive pricing monitor that scrapes three competitor pages and emails you when something changes
  • Market news digest filtered to your specific industry, delivered every morning

Brief to adapt:

THE PROBLEM: I spend 45 minutes every Monday pulling numbers from three
different spreadsheets to write my weekly summary email.

INPUT: Three Google Sheets exports saved as CSVs in /data/weekly/

OUTPUT: A formatted email draft in /output/ with: revenue vs last week
(absolute + %), top 3 customers by revenue, any new customers,
any customers who haven't ordered in 30 days.

TRIGGER: I run it manually, Monday mornings.

CONSTRAINTS: Never send automatically. Create a draft only.

SUCCESS: I can paste the output directly into my email with no edits.

2. Meeting Operations

What it is: Automating the before and after of every meeting — preparation and follow-through.

What Fernandez built: A /meetsync command that automatically organizes meeting transcripts, files them by client, and extracts action items. A hiring workflow that takes a resume paste and returns: pipeline update, interview questions tailored to the team's current gaps, and screening call prep. All triggered by simple commands he runs himself.

What you might build:

  • Pre-meeting brief: paste a name, get back context on the person, your last conversation, and suggested questions
  • Post-meeting processor: paste raw notes, get structured action items with owners and deadlines
  • 1:1 tracker: paste what was discussed this week, maintain a running record of each direct report

Brief to adapt:

THE PROBLEM: After every client call I have unstructured notes that I
never organize properly. Action items fall through.

INPUT: A text file I paste into a folder called /meetings/raw/

OUTPUT: A structured markdown file in /meetings/processed/ with:
- 3-line summary
- Action items (owner, deadline, description)
- Key decisions made
- Questions that need answers

TRIGGER: I drop a file into /meetings/raw/ and run the command.

CONSTRAINTS: Don't infer things not in the notes. If a deadline isn't
mentioned, put TBD. If an owner isn't named, flag it.

SUCCESS: I can share the output directly with the meeting participants.

3. Data Processing and Cleanup

What it is: Making messy data clean, or combining data from multiple systems into something readable.

What enterprises are doing: Anthropic reports Claude Code being used at Newfront insurance to automate HR administrative work — reclaiming over one month per year in administrative time. The NYSE is using it to automate engineering workflows from Jira tickets. The tasks are different in scale, but the pattern is the same: messy input, structured output.

What you might build:

  • Invoice parser: drop PDFs in a folder, get a clean CSV with amounts, vendors, dates
  • CRM cleanup: take a messy export, deduplicate contacts, standardize phone/email formats
  • Expense categorization: take a bank statement export, categorize by type, flag anything over $500

Brief to adapt:

THE PROBLEM: Our sales team sends me pipeline updates in three different
formats (Excel, Google Sheets export, and a weekly email copy-paste).

INPUT: Files in /data/pipeline/ (mixed formats, whatever arrives that week)

OUTPUT: One clean CSV in /output/ with columns: Company, Stage, Value,
Owner, Last Contact, Next Step, Days Since Update.

TRIGGER: I run it manually, Friday afternoons.

CONSTRAINTS: Flag any row where "Last Contact" is more than 14 days ago.
Don't delete the original files. If a field is missing, leave it blank
rather than guessing.

SUCCESS: One clean file I can review in 10 minutes instead of 45.

4. Prototype Validation

What it is: Building a rough working version of an idea to test it before involving your team.

What Siemiatkowski does: He builds prototypes in ~20 minutes to validate ideas before any engineering discussion. He goes into those conversations with a working prototype, not a description. The engineers know exactly what he means. Edge cases are already visible.

This is Claude Code's highest leverage for an executive: not building production systems, but making requirements concrete before handing them off. One hour of prototyping produces a better developer brief than one hour of requirements meetings.

What you might build:

  • A rough dashboard showing data from a CSV to test whether the visualization is what you actually want
  • A simple web form to test a workflow before asking engineering to build the real version
  • A script that simulates a business process to see what breaks before it's live

Brief to adapt:

THE PROBLEM: I want to test whether a weekly "customer health score"
dashboard would be useful before I ask engineering to build it properly.

INPUT: A CSV export of our customer data (attached)

OUTPUT: A simple local web page that shows:
- A table of customers sorted by health score (high = green, low = red)
- Health score = orders in last 90 days × average order value ÷ days as customer
- Click a row to see order history

TRIGGER: I open it in my browser locally. No hosting needed.

CONSTRAINTS: This is a prototype. It doesn't need to be pretty or fast.
It needs to show me whether the concept is right.

SUCCESS: I can share the concept with my head of engineering and they
understand immediately what I want.

5. Work Logging and Knowledge Management

What it is: Building personal systems to capture and retrieve what you've done, decided, and learned.

What Stockton built: A structured work logging system that captures dates, categories, subjects, summaries, and related files — maintained through simple commands. The insight: "If you capture what you did in a structured way, Claude can use that information later." The log becomes queryable institutional memory.

What you might build:

  • Decision log: record every significant decision with the context, alternatives considered, and outcome
  • Client relationship tracker: maintain a running record of every client interaction, indexed by company
  • Weekly review automation: pull your calendar and task log into a structured weekly review template

Brief to adapt:

THE PROBLEM: I make dozens of decisions a week and lose track of why
I made them. When things don't work out, I can't trace back the reasoning.

INPUT: A short note I dictate or type after each significant decision:
who was involved, what we decided, what the alternatives were,
what we'll know by when to evaluate whether it was right.

OUTPUT: A markdown file in /decisions/ named with today's date,
plus an updated index file that lists all decisions in reverse
chronological order with one-line summaries.

TRIGGER: I run a command, paste my note, it formats and files it.

CONSTRAINTS: Don't change my wording. Format it, don't edit it.

SUCCESS: In 6 months I can search /decisions/ and understand
why I made any decision.

6. Hiring and Team Operations

What it is: Automating the repetitive cognitive work around hiring, performance, and team management.

What Fernandez built: A hiring workflow that takes a resume and returns: an updated pipeline entry, interview questions based on the team's current gaps, and screening prep tailored to the candidate. The information is there — it just needed structuring. Claude Code does the structuring.

What you might build:

  • Resume screener: paste a resume, get back a one-page assessment against your criteria
  • Interview prep: given a candidate's background and your job criteria, generate focused questions
  • Performance summary: given your notes from a quarter of 1:1s, generate a structured review

Brief to adapt:

THE PROBLEM: I interview 5-10 candidates a month and prep for each
interview differently every time. I want a consistent, useful brief.

INPUT: A candidate's resume or LinkedIn URL, pasted into the terminal.

OUTPUT: A one-page interview brief:
- 3 sentences on their background
- 3 strengths I should probe to verify
- 3 gaps or questions I should address
- 5 specific interview questions based on what I see

TRIGGER: I run it manually before each interview.

CONSTRAINTS: Don't generate generic interview questions. Base them on
what's actually in the resume. Flag anything that seems inconsistent.

SUCCESS: I feel prepared after 5 minutes, not 30.

The Pattern Underneath All of These

Every CEO use case follows the same shape:

  1. Repetitive cognitive work — something you do regularly that follows a pattern
  2. Messy inputs — data from multiple places, in multiple formats
  3. Structured outputs — something you can act on directly
  4. Your judgment stays in the loop — Claude Code creates the draft, you review and decide

The tools that work best are the ones where you remain in control of the decision while Claude Code handles the assembly. The tool shouldn't decide. It should organize, format, and prepare — so your time goes to the part only you can do.

Further Reading

Next: Reading Code You Can't Read

Stay in the loop

Occasional updates on AI systems, tools, and new writing.

Ormus — Diego Bodart

Au