Get the deck and skill pack

Enter your email to view the WSC CEO workshop materials.

1 / 19

The Operator-Engineer

How CEOs turn AI capability into time back, better workflow design, and safe operating leverage.

Sam Gaddis & Matthew Hall  ·  Runpoint  ·  WSC Webinar  ·  May 2026
  • Hold for ~15 seconds while people settle in
  • "I'm Sam Gaddis, and I'm here with Matthew Hall. We run Runpoint together. We build software for companies across manufacturing, cybersecurity, energy, financial services, and more."
  • "Today I want to talk about a shift that's happening right now — the people who understand the business are becoming the people who build the tools, and that changes everything."
  • "This session is for CEOs. You already have the hard part: deep operational knowledge of a real business. The question is how to turn that knowledge into leverage without adding another giant project to your plate."
  • "We'll keep the framing short, show you what the high end looks like, and spend the working portion on something useful tomorrow: an assistant for email, calendar, and follow-up."
  • NEXT: "Start with the actual constraint in the room."

The build is no longer
the bottleneck.

What most companies ask

Can AI write the code? Can it build the thing? That question is already behind the frontier.

What now matters

Knowing what to build, whether it is safe to use, where the data lives, and which workflow should ship first.

AI did not make every company instantly faster. It made operational judgment the scarce resource.

  • The hard part is not typing code anymore. The hard part is choosing the right business problem and putting guardrails around the answer.
  • "This technology can touch finance, ops, customer support, sales, compliance, and delivery. That is why prioritization is suddenly an executive skill."
  • "Most companies will do the same work a little faster. The winners will redesign the work itself."
  • Anchor the talk: "The person who can connect business taste, technical acumen, and AI intensity is the new leverage point."
  • NEXT: "The bottleneck moved."

The Bottleneck Moved

Quoted excerpt

“God made men differ in strength, but Sam Colt made them equal. God made men differ in intelligence, but Sam Altman made them equal.”

When intelligence gets cheaper, taste and judgment matter more.

Adapted from tomie (@tomieinlove), X, May 20, 2026.
  • Use this as the cleaned-up version of the online quote. Keep the Colt/Altman analogy and skip the first sentence from the screenshot.
  • "The useful version is simple: once intelligence gets cheaper, taste matters more."
  • "That is why CEO judgment matters. The model can produce output. It still needs someone who knows what should exist."
  • NEXT: "Place that judgment on a simple matrix."

The AI Competence Matrix

Uses AI Does not use AI
Slop Cannons Lots of output. None of it good.
Turbo Brains Fast, correct, compounding value.
Dead Weight Slow and wrong.
Steady Hands Reliable, but limited throughput.
Bad judgment Good judgment
  • Walk each quadrant. Start bottom-left, go clockwise
  • Pause on the arrow: "AI doesn't fix judgment, it amplifies it."
  • Turbo Brains = where operator-engineers live. These are the people to invest in — through hiring, through internal development, through giving them room to build.
  • Key line: "AI moves people up the chart. The question is which column they're in."
  • NEXT: "First choose the operating mode."

The Four Ways to Work with AI

Most CEO workflows should start in sparring or centaur mode. Agent mode comes later, after the judgment boundary is clear.

Mode 01

Artisanal

100% human / 0% AI

You do the work yourself because quality, context, or trust matters more than speed.

Cost: leverage left on the table.
Mode 02

Sparring

80% human / 20% AI

AI reflects, critiques, pressure-tests, and helps you see around corners.

Cost: easy to mistake feedback for work.
Mode 03

Centaur

40% human / 60% AI

You set taste and direction. AI produces the draft, artifact, or analysis.

Cost: review fatigue.
Mode 04

Agent

10% human / 90% AI

You frame the job, constrain the tools, and review the result after the work runs.

Cost: surrender in domains you care about.
Adapted from Jonny Miller (@jonnym1ller), “The Four Operating Modes,” X, May 13, 2026.
  • This slide gives the workshop language for tool use. The CEO assistant starts as sparring and centaur, not agent mode.
  • "Artisanal is doing it yourself. Sparring is using AI as a thinking partner. Centaur is where the model produces the draft and you steer. Agent is delegation."
  • "Most of the room should not jump straight to agent mode. The first assistant should draft and summarize while the CEO keeps approval authority."
  • NEXT: "That judgment has a role attached to it."

Autonomous Proposals

Live Demo

One Slack message. CRM updated. Proposal drafted. Human reviews before it goes out.

Slack Claude Code Granola Runpoint CRM (Airtable) Gmail
sam → #runpoint-ops
@hermes look at my last call, update the CRM, and create a proposal draft in my email
1

Read the call transcript

Pulls the Granola transcript from the most recent call. Extracts scope, stakeholders, timelines, and open questions.

2

Update the CRM

New contacts added. Opportunity stage advanced. Notes attached to the right account. All in Airtable.

3

Draft the proposal

Matches scope to our past work, pulls pricing from the rate card, generates a full proposal in our template.

4

Email the draft to me

Lands as a reply-ready draft in Gmail. I tweak and send. An afternoon of work becomes 10 minutes of review.

1 Slack message  ·  4 autonomous steps  ·  afternoon → 10 minutes of review
  • This is the one that broke my brain. Built it a couple weeks ago — haven't touched a proposal the old way since.
  • "The whole flow runs on one Slack message. Four things happen autonomously — it reads my call transcript, updates the CRM, drafts the proposal in our template, and emails me the draft."
  • Before: "An afternoon of clicking through Airtable, summarizing the call, opening the template, finding pricing, writing the proposal."
  • After: "10 minutes of review and tweaks on a draft that's 90% there. The AI does the draft work; a human still decides what gets sent."
  • Key point: "This isn't the build I'm most proud of — it's the one I use most. That's the point. Operator-engineers build the tools they actually use."
  • Live demo if network cooperates: send the Slack command live, then show the CRM update and email draft a minute later.
  • "This is not a pitch for an engagement model. It is proof that the boring work around a CEO can be redesigned."
  • NEXT: "Now we can name the kind of person who builds this."

The Operator-Engineer

The operator-engineer sits at the intersection of business taste, technical acumen, and AI intensity.

Business
Taste
Technical
Acumen
AI
Passion
Operator-Engineer
Knows which problem is worth solving
Comfortable at the code level, not just the prompt level
Ships fast, iterates faster, throws away freely
Treats AI as a power tool, not magic
Thinks in systems, not features

Any two of three makes you useful. All three makes you dangerous.

  • The Venn is the core idea. Walk through missing-circle combos:
  • Business taste without technical chops = great specs nobody builds
  • Technical acumen without AI = steady hand, reliable but slow
  • AI passion without business taste = slop cannon
  • You need all three. Traits on the right are what it looks like in practice
  • "Any two makes you useful. All three means you can go from problem to shipped solution without waiting on anyone else."
  • NEXT: "The safety model follows from that."

You do not have enough hours.

We are here to use AI for something concrete: giving the CEO enough time back to think about the larger strategic work.

1

Email keeps fragmenting the day

The assistant should find what needs a reply, draft the response, and leave the send decision with you.

2

Calendar hides the real work

The assistant should turn meetings into prep notes, follow-ups, and a short list of decisions that need CEO attention.

3

Follow-up leaks between systems

Most missed leverage is not a bad strategy. It is a task, promise, or waiting item that never made it into the operating rhythm.

4

Delegation needs cleaner packets

A good assistant turns loose thoughts into context, owner, deadline, constraints, and a clear definition of done.

Before the transformation work, build the tool that makes tomorrow less crowded.

  • Use this to align with Will's read from the call. These are CEOs, and the shared constraint is time.
  • "If you're already out of hours, a six-month AI transformation sounds like one more thing you do not have time for."
  • "So we are going to start smaller. Email, calendar, follow-up, delegation. Boring surfaces. Very high leverage."
  • "The point is not to replace judgment. It is to protect enough of your day that judgment can show up."
  • NEXT: "There is still a full spectrum of AI work. I want to show both ends."

Build a CEO Assistant

The first skill should help with the two places time leaks fastest: email and calendar.

D

Daily brief

Today's meetings, prep notes, open loops, and the emails that need a response before they become someone else's blocker.

R

Reply drafts

Drafts for the emails you should answer yourself, plus a delegation packet for messages someone else should own.

W

Weekly

What happened last week, what is coming next week, who is waiting on you, and where the company needs a CEO decision.

It drafts and summarizes. You approve sends, cancellations, commitments, and promises.

  • This is the live build. Keep it tool-agnostic enough that the room does not get stuck in vendor setup.
  • "We are going to give you a prompt you can copy, then change a few fields: company context, voice, tools, and approval rules."
  • "If Gmail access works, great. If Outlook needs an admin, we can still build the routine with copied email and calendar text. The workflow is the point."
  • "The safety rule is simple: summarize, draft, and queue. Do not send or make external commitments without human approval."
  • NEXT: "The CEO still has a job in making this usable."

Your Mad-Lib Setup

Paste the prompt as Project Instructions. Add the priorities file to Project Knowledge. Fill the [BLANKS].

System Prompt

# Role
You are [YOUR NAME]'s executive assistant.

# Goal
Save me time and keep things from slipping. Draft work, surface what matters, never act without approval.

# Where to find what you need
- priorities.md (Project Knowledge) — who I am, what matters this quarter, who matters most, what to flag, what to ignore, how I talk.
- Today's calendar and inbox via the connector.
- Uploaded transcripts in Project Knowledge, if present.

Read priorities.md before every response.

# Skills and routines
- Daily Digest — weekdays 7am. Brief + Gmail drafts.
- Friday Recap — Fridays 3pm. Week recap + next-week plan.
- Ad-hoc: triage, draft a reply, prep a meeting, summarize a thread. Use priorities.md to decide what matters.

# Rules
1. Draft, never send. No external commitments without me.
2. Never invent facts or commitments. Flag for confirmation.
3. Cite sources by name (email subject, meeting title, doc).
4. Follow the voice block in priorities.md.
5. When in doubt, ask.

priorities.md project knowledge

# My priorities

## Who I am
- [TITLE] at [COMPANY]
- [ONE LINE: WHAT THE COMPANY DOES]
- [ONE LINE: WHO WE SERVE]

## What matters this quarter
1. [TOP OUTCOME #1 + target date]
2. [TOP OUTCOME #2]
3. [TOP OUTCOME #3]

## Who matters most
- [NAME] ([ROLE]): [WHY, ONE LINE]
- [NAME] ([ROLE]): [...]
- [NAME] ([ROLE]): [...]

## Always flag
- Anything from [NAMES]
- Anything mentioning [DEAL / TOPIC]
- Invoices over $[AMOUNT], contracts, anything legal
- Threads I owe a reply > [N] days

## Do NOT surface
- Newsletters and marketing
- Calendar invite spam
- Automation noise (only humans @-mentioning me)
- Receipts under $[AMOUNT]

## Voice
- Tone: [DIRECT / WARM / FORMAL]
- Lead with the punchline. 3 things; demote the rest.
- Short bullets. Headers when there's structure.
- Every action item has an owner and a date.
- Never use: [3-5 WORDS YOU HATE]
- Estimates: complexity (S/M/L), not time.
- "I don't know" is fine.

The [BLANKS] are the lever. Sharpen them, and the assistant gets noticeably better in one edit.

  • Live moment. Have the room open claude.ai or ChatGPT Projects in another tab. Walk them through pasting the System Prompt as Project Instructions, then creating priorities.md in Project Knowledge.
  • "The Markdown headings are not decoration. They're section labels. The model reads them as 'these are distinct rules.' You can rename them. You cannot skip them."
  • "The first edit you should make is the voice block. Add three words you never want to read in your own drafts. That single change is the highest leverage edit in this whole setup."
  • NEXT: "Now the two routines that turn it into an EA, not a chatbot."

The Two Routines

Schedule these in Claude Co-work (Routines) or ChatGPT Projects (Scheduled Tasks). Drafts land in your inbox. You approve sends.

Daily Digest weekdays · 7:00 am

It's the start of a workday. Produce my Daily Digest using priorities.md, today's calendar, and unread email.

## Settings (edit these)
- POSTURE: balanced — drafting aggressiveness. One of: aggressive (draft anything plausible, including first-pass replies to high-priority emails marked as such), balanced (default — draft scheduling/acks/known answers/polite declines), conservative (draft scheduling and acks only; suggest the rest).
- MAX_EMAILS: 7 — cap on emails surfaced in Inbox.

## Today
3 lines max: the most important thing, the most-likely to derail, one optional opportunity worth 30 minutes.

## Meetings
Every meeting today, chronological. For each: time, title, attendees, 1-line prep, link if present.

## Confirmed scheduling (propose only — never create without my approval)
Scan replies in the last 24-48 hours to scheduling drafts I sent. For each confirmation: who confirmed, the time they picked, the meeting topic, proposed event title, attendees. Counter-proposals or unclear replies stay in Inbox.

Show me the list. Wait for me to say "add it" — or specify which ones — before touching my calendar. If I approve and you have calendar-write, create as tentative events with attendees. If not, give me the details to add myself.

## Inbox
Surface up to MAX_EMAILS emails that need me today, grouped below. Respect POSTURE for the Drafted/Suggested split.

### Drafted (in Gmail, ready to send)
Per POSTURE, always draft these without asking:
- Scheduling requests (propose 2-3 times from my calendar, respect priorities.md's protected blocks).
- Answers I've given before to similar questions.
- Acknowledgments, "got it," "thanks, will review."
- Polite declines for things I don't do.

If POSTURE is aggressive, also draft first-pass replies to high-priority emails; label them "[aggressive draft — review carefully]".
If POSTURE is conservative, draft only the first two bullets; push the rest to Suggested.

For each: subject, sender, 1-line of your draft, link.

### Suggested (don't draft yet)
High-priority emails where you have a strong opinion but should not act without me. Propose 2-3 sentences each. I'll say "draft it" or rewrite.

### Waiting
I owe someone, the thread has gone quiet, or a reply is overdue. Include last-activity date.

### Skim
Worth knowing. No action. One line each.

## Slipped
Anything from priorities.md or yesterday that should have moved and didn't. Call me out.

## Patterns
Email types or senders repeating this week (3+ in 7 days). Examples: "3rd scheduling request from [NAME]", "5th 'how do I X' matching a pattern." For each pattern, propose one line for priorities.md that would let you draft confidently next time.

## Also
2 lines max of small things.

Voice: operator-grade. No AI-speak. Lead with the punchline. Cite sources by name.

Friday Recap fridays · 3:00 pm

It's Friday afternoon. Produce my weekly recap and next-week plan using priorities.md, this week's calendar, and this week's email. Include transcripts if present.

## Settings (edit these)
- INITIATIVE_COUNT: 3 — how many initiatives to defend time for next week.
- BLOCK_SIZE: 2-3 hours — preferred deep-work block duration.

## First, ask me two things and wait for my answer
1. Which initiatives do I want to defend time for next week? Propose INITIATIVE_COUNT candidates based on priorities.md and what slipped this week, then confirm.
2. Anything personal, travel, or off-calendar I should plan around? Examples: travel days, family commitments, focused-writing time I haven't booked, doctor appointments.

Use my answers to shape the Next Week and Time Blocks sections below.

## Week in one paragraph
What the week was about. Not a list. Lead with the most consequential thing.

## Decisions locked
Every decision made this week. Name it, date it, rationale. If unconfirmed, label "confirm with [person]".

## Slipped
Tasks, replies, follow-ups that didn't happen. No softening.

## Open loops
Threads where I owe someone. For each: who, what, by-when.

## Next week top initiatives
INITIATIVE_COUNT initiatives to defend time for, aligned to my answer above. For each: owner, real by-when date, success criterion.

## Calendar heads-up
Next week: conflicts, important first meetings, missing prep.

## Time blocks (propose only — never create without my approval)
Propose calendar holds next week for each top initiative. Respect priorities.md protected blocks and the personal/travel items I named. For each block: title prefixed with [HOLD], real slot from open calendar, BLOCK_SIZE where possible (smaller only if the calendar forces it), description naming the initiative and what done looks like.

Show me the proposed list. Wait for me to say "create them" — or specify which ones — before touching my calendar. If I approve, create as tentative events. If you don't have calendar-write access, give me the list to drag in.

## One question
The single question I should sit with this weekend. A question, not advice.

Run each one once now. If the output is off, edit priorities.md. That's where the lever lives, not in the routine prompt.

  • Live moment. Have everyone hit "Run now" on the Daily Digest. Expect a brief that references real calendar + inbox. If the connector is blocked, paste 3-5 recent emails inline and run it anyway. The workflow is the point.
  • "The output won't be perfect on day one. Don't chase the prompt. Sharpen priorities.md instead. Add the name you noticed it missed. Strike the word that slipped through your voice block."
  • "Friday Recap will be thin the first week. It compounds. Tell them that up front so nobody bails after one run."
  • NEXT: "The CEO's job is to clear the path so this actually fires."

Clear the Path

The assistant only works if the workflow has context, access, and a clear approval line.

1

Give it context

Your role, company, priorities, tone, direct reports, meeting rhythm, and the decisions only you should make.

2

Give it bounded access

Email and calendar are enough for the first version. Add CRM, notes, and meeting transcripts only after the base routine is useful.

3

Set the approval rules

Drafts are fine. External commitments, customer messages, contracts, payments, and calendar changes need human approval.

4

Name the next owner

Find the person who can improve the workflow after today. The CEO should use the assistant, not become the help desk for it.

The CEO should define the judgment boundary. The assistant should handle the draft work inside it.

  • This keeps the workshop from becoming tool theater. The skill needs context and boundaries before it needs more integrations.
  • "The most important setup is not technical. It is deciding what the assistant is allowed to do and where it has to stop."
  • "Start with bounded access. Email and calendar are enough. Then add meeting notes, CRM, or project systems once the routine earns trust."
  • "If this becomes useful, hand the next iteration to the operator-engineer in your company. You probably already know who that is."
  • NEXT: "Once they know how to build one skill, give them a menu."

The CEO Skill Pack

After the assistant, the next step is a library of copyable skills for recurring CEO work.

7

skills to copy, edit, and run

A clever demo fades. A reusable skill becomes a habit your company can keep improving.

  • This comes directly from the workshop plan in the transcript. Give them the assistant live, then a menu of other skills behind the email wall.
  • "Once you understand the pattern, the assistant is just the first example."
  • "The pack should include meeting prep, follow-up, delegation, finance questions, customer risk, vendor review, board update drafting, and workflow mapping."
  • "They can grab the deck and the skills afterward, then bring us in when the workflow becomes bigger than a prompt."
  • NEXT: "Make the setup simple enough that people actually get through it."

What You Can Do This Week

1

Run the assistant for five days

Use the daily and weekly routines before you add integrations or more ambitious workflows.

2

Pick one company workflow

Choose the recurring process people complain about most: finance review, customer follow-up, vendor review, onboarding, or meeting follow-through.

3

Find the operator-engineer

Look for the person who already built the spreadsheet, automation, dashboard, or internal tool everyone depends on.

  • "Do not leave with a list of twenty AI ideas. Leave with one routine you will actually run."
  • "After five days, you will know whether the assistant is saving time or just producing more text."
  • "The next workflow should be something painful and repetitive. Let the result sell the next one."
  • "The operator-engineer is probably already in your company. Your job is to notice them and clear a path."
  • NEXT: "Recap the workshop build steps."

Build the CEO Assistant.
Then make it useful.

The workshop was not about installing a tool. It was about creating one small assistant that knows your priorities, reads the right systems, drafts the routine work, and improves each week.

1

Create the project

Start a Claude/Cowork project for your EA, give it a name, turn memory on, and begin with one clear sentence about who it serves.

2

Connect the basics

Add email, calendar, and Drive or Microsoft 365. If admin approval blocks a connector, use copied email and calendar text first.

3

Paste the mad-lib

Use the project instructions to define the goal, available context, approval rules, writing style, and where the assistant should look.

4

Add priorities

Create a priorities file with your company, current quarter, key people, alert rules, voice, and anything the assistant must flag.

5

Schedule two routines

Run a daily digest for meetings, inbox triage, and reply drafts. Run a Friday recap for next week, blockers, follow-ups, and focus time.

6

Coach it weekly

Correct mistakes, start fresh chats inside the project, review memories, and add transcripts, CRM, or tasks only after the base loop works.

Start with one low-blast-radius loop: inbox + calendar become brief + drafts. You approve sends, commitments, and calendar changes.

  • Source-backed from the workshop transcript. Matthew walked the room through creating a Claude/Cowork project, connectors, a priorities file, and scheduled daily/weekly routines.
  • "The goal is not a perfect autonomous EA. The goal is one useful loop that runs every day and gets corrected as it learns your business."
  • "Keep the assistant in draft-and-summarize mode until the routine earns trust."
  • "Once the basic loop works, add more context: transcripts, CRM, todo lists, or Slack."
  • NEXT: "Download the materials and keep iterating."

Start with personal leverage.
Then widen the surface area.

There are two useful conversations: the assistant you can build today and the AI-native operating system that takes months.

Tomorrow
A CEO assistant for email, calendar, prep, replies, follow-up, and delegation
to
Months
AI-native operating systems across sales, finance, operations, delivery, and customer work

We will spend the workshop on the left side. The demos show what the right side can become.

  • This is the framing from the call. We can mention the larger AI-native systems, but the workshop should focus on what helps tomorrow.
  • "On the left side, we are building a practical assistant that gives you time back. On the right side are the deeper operating systems we build with companies."
  • "A transformation like the right side is usually a real company project. It involves data, rollout, permissions, and behavior change."
  • "But the left side is something we can do in the room."
  • NEXT: "Here is what the right side looks like when it works."

What makes this safe to use?

Do not ask AI to be trusted. Design the workflow so trust is earned in narrow, observable steps.

1

Human approval before external action

AI drafts, updates, recommends, and queues. A person approves emails, payments, contracts, customer messages, and irreversible changes.

2

Permissioned tools, not open-ended agents

Give the system scoped functions: read this CRM record, draft this proposal, flag this exception. Do not give it the keys to the business.

3

Audit trails and reversibility

Every action should be logged, attributed, diffable, and easy to undo. The safety layer is operational, not philosophical.

4

Start with low-blast-radius workflows

Begin with drafts, summaries, reconciliations, and internal queues. Move toward autonomy only after accuracy is measured in production.

Safe AI is not a model choice. It is a workflow design choice.

  • This is the credibility slide. CEOs will be asking this even if they do not say it out loud.
  • "The way to use AI safely is not to have a religious debate about whether models hallucinate. They do. Design around it."
  • "Human approval before external action. Scoped tools instead of unlimited agency. Audit trails. Reversibility. Low-blast-radius starting points."
  • "If you follow those rules, AI can be incredibly useful before it is fully autonomous."
  • NEXT: "Now let's build the low-blast-radius version together."

Same pattern.
Different businesses.

Operator-engineers do not build generic AI toys. They turn specific operating pain into shipped workflows.

Runpoint OS

  • Sales pipeline
  • Proposal generation
  • Recruiting workflow
  • Forecasting and capacity
  • Used by the whole team

Austin Custom CNC

  • AR / AP visibility
  • P&L cleanup
  • CRM and project tracking
  • Contract analysis
  • Weekly operating cadence

This Deck

  • Call transcript to outline
  • Synced presenter notes
  • PDF and OG automation
  • Single HTML artifact
  • Built as the need appeared
Internal services  ·  manufacturing  ·  sales workflow  ·  presentation tooling

The reusable pattern is not the app. It is the habit: find painful workflow, expose data, build a narrow tool, measure, repeat.

  • Use this as a proof wall, not a demo queue. Do not walk every feature. Pick one concrete example from each if needed.
  • "Runpoint OS is our internal version. Austin Custom CNC is a portfolio-company-style operating system. This deck is a tiny meta example."
  • "The important thing is not that the apps look similar. The important thing is that the operating motion is similar."
  • "Once you learn to turn workflows into skills, the next business function gets easier."
  • NEXT: "Here is a smaller version of the same pattern that a CEO immediately understands."
  • ~10 minutes for Q&A
  • Frame this as the post-gate resource page. They scan once, enter email once, then download the skill pack and PDF here.
  • If people still need the URL, point to the QR code on the right.
  • If questions are slow: "What's the most painful workflow in your organization that you wish you could automate tomorrow?"