Social Media Mastery 2026, with Mr. Grateful

Scale without
losing your taste.

Session one. AI systems for content velocity and testing.

Before anything else.
Thank you for being here.

The mastery, end to end

Four sessions with me.

01
Velocity and testing, May 21
02
Install one system live, May 26
03
Scale and integrate, July 2
04
Strategy sprint, July
01

Today.

This session is content and demonstration. On the 26th, we build together.

What we cover
  1. 01The problem
  2. 02The intelligence layer
  3. 03AI Intuition
  4. 04The Creator Content System
  5. 05Build a skill
Two tools today
Claude
Claude
Assistant, coworker, and coder.
Creator Content System
Creator Content System
A tool built just for you.
The running example

I'm building a new page
live, alongside you.

GratefulFuture, created with Sentient, the agency running the largest tech network on social, reaching more than a hundred million followers. Everything today happens on this real page.

instagram.com/gratefulfuture

iPhone screenshot
The good news

You have the foundation. Audience. Message. Platform. You are already making content.

The process is manual.
And it is heavy.

You don't need more discipline.
You need a system that runs
when motivation doesn't.

The other end of the spectrum

Hand the whole thing to AI
and you get slop.

Example of AI slop
So the work becomes

Raising your taste and discernment, and learning the skill of instructing AI. AI makes better decisions faster, so your enjoyment goes up, not your output of slop.

By the end of today

Identify one repetitive part of your workflow. Then design a simple AI-assisted process for it.

How Claude works with you
Assistant
Manual help, on demand.
Coworker
Work that runs on a schedule.
Coder
Custom tools built for you.
Already built, with skills and connectors
Research thread
Research
Ground a topic before you speak.
Scripting thread
Scripting
Draft in your voice from context.
Higgsfield image
Content
On-brand visuals.
Page in code
Web
A page or tool, in code.

A new layer just appeared
between you and your work.

Where you stand now
Reach Script Research Image Web Schedule Intelligence Layer The model AI holds of your company You purpose · taste · discernment Reality
What rises in value

Three things only you can hold.

Purpose
Why the thing exists at all.
Taste
How it should exist.
Discernment
Where AI does and does not belong.

AI Intuition.

02

How to apply it.

A simple way to decide, step by step, where AI belongs.

Tag every step by skill and enjoyment
Good at it
−   +
Collab with AI
You can, but would rather not. Steer it; let AI do the reps.
+   +
Do it yourself
Your craft, your zone. Keep it human.
−   −
Automate
Hand it off fully. Let the system run it.
+   −
Learn
You enjoy it but aren't there yet. Grow in.
Dislike Bad at it Enjoy
Tag each step. Here's why it helps
  1. It isolates the variables in your system.
  2. It measures how much is AI versus human.
  3. It flexes to your strengths and what you enjoy.
The canvas you'll tag

The Creator
Content System.

Your whole pipeline on one map. Idea, through publishing, to what the numbers teach you. Built for an Instagram-first creator who tests, then repurposes.

domenicashburn.com/lab-tools/content-system

Creator Content System
03

Now, use AI.

Start with the simplest possible move.

The simplest move

Pick one step. Just ask AI for help with it. No setup. See what raw capability gives you.

With no context, structure, skill,
goal, or schedule, raw AI just adds
another layer of manual work.

The architecture that fixes that
PROJECT A standing context with a mission Instructions Your standing intent Knowledge Your docs, uploaded once Chat The work happens here Skills The repeatable how Connectors Live reach to other systems Artifacts What leaves the room Ship
Your first move in Claude
  1. 01
    Create a project.
    One home for the whole pipeline.
  2. 02
    Goal as the instructions.
    The standing intent every chat inherits.
Project setup
Live, leave the slides

Build the project.
Ask again, inside it.

Set the goal as instructions, add the brand docs, then run the same step. Same ask, different result.

04

Skills.

When a step repeats, stop re-explaining it.

What it is

A skill is the repeatable how. Claude reads it and knows on its own when to use it. Build one when a step repeats and you want it done by your guidelines.

What's actually in one

A folder. One skill, one job.

SKILL.md
The instructions, written in plain language
Description
The one line that tells Claude when to use it
References
Deeper notes, pulled in only when needed
Scripts
Code for the steps that never change

Claude reads only the description until a task fits. The rest loads when it is needed.

From your 31 skills, the ones that matter here
Research
Dual-lens research, ground a topic before you speak
Script
Response drafter, writes in your voice from context
Image
Higgsfield images, on-brand cinematic visuals
Web
Keynote builder, a doc becomes a presentation
Loop
Daily X post, research, image, thread, post in one
Meta
Skill builder, builds your next skill
The build loop
  1. 01
    Intent.
    What it does, when it triggers, what it outputs.
  2. 02
    Description.
    The one line that decides whether it ever fires.
  3. 03
    Draft the skill.
    A confident playbook, not an encyclopedia.
  4. 04
    Test, then iterate.
    Run it on real prompts. Refine. Repeat.
When you iterate, look for
  1. Read the transcript, not just the final output.
  2. Cut any step that isn't pulling its weight.
  3. Add the why wherever it went off the rails.
Judgment

Reading the situation.

Writing in your voice, choosing what to say. Give it intent. No fixed rule survives real variety.

Rote

The same every time.

Formatting, date-stamping, posting at a set hour. Give it a rule or a script. You want zero variance.

Two kickoff templates. Copy and fill the brackets
Short 
I want to build a Claude skill with you for [STEP] of my content system.
Right now I do this by hand: [WHAT YOU DO TODAY].
Done looks like: [THE OUTPUT IN ONE SENTENCE]. It matters because [WHY].
Ask me whatever you need, one question at a time. Then draft a lean SKILL.md
that explains the why behind each step so it handles cases I did not list, and
tell me which steps should be fixed scripts and which need judgment. Then write
the description so it triggers reliably, and test it with me against a few
realistic near-miss cases.
Long 
I want to build a Claude skill with you. Do not write any files yet. First help me
get the design right, then we draft, then we optimize the trigger.

THE ASK
Build a skill that handles [STEP] of my creator content system. This is the part
where I currently [WHAT YOU DO BY HAND TODAY, IN PLAIN WORDS].

COMMANDER'S INTENT
- Done looks like: [THE FINISHED OUTPUT, AS ONE SENTENCE. WHAT LEAVES THE ROOM.]
- I will know it is good when: [HOW YOU JUDGE QUALITY. THE BAR.]
- This exists so that: [THE PURPOSE BEHIND IT. WHY IT MATTERS. THIS LETS YOU
  REASON THROUGH EDGE CASES I DID NOT LIST, INSTEAD OF FOLLOWING RIGID RULES.]

WHEN IT SHOULD RUN
- Trigger situations: [THE MOMENTS THIS SHOULD FIRE. WHAT I AM DOING OR SAYING.]
- Inputs I will hand it: [WHAT I GIVE IT. TRANSCRIPT, TOPIC, FILE, URL, RAW NOTES.]
- Output it produces: [THE DELIVERABLE. DRAFT, THREAD, CAPTION SET, PROMPT PACK.]

THE STEPS
[LIST THE STEPS THE SKILL SHOULD HANDLE, IN ORDER. BE A LITTLE TEDIOUS.]
1. [STEP]
2. [STEP]
3. [STEP]

JUDGMENT VS ROTE
- Steps that need judgment (give intent, let yourself reason): [WHICH STEPS REQUIRE
  READING THE SITUATION. WRITING, CHOOSING, SHAPING.]
- Steps that are the same every time (make these a fixed rule or script): [WHICH STEPS
  ARE DETERMINISTIC. FORMATTING, NAMING, ORDERING, COUNTING.]

VOICE AND HARD RULES
- My voice: [SHORT DESCRIPTION OR "MATCH MY USUAL STYLE".]
- Non-negotiables: [BRAND RULES THAT MUST NEVER BREAK. E.G. NO EM DASHES, NO
  EXCLAMATION MARKS, AI STAYS INVISIBLE, NEVER PROMISE OUTCOMES.]

BOUNDARIES
- Out of scope: [WHAT THIS SKILL SHOULD NOT TOUCH.]
- Stays with me: [ANY STEP THAT IS IRREVERSIBLE, PUBLIC, OR HARD TO CHECK. POSTING,
  SENDING, SPENDING. THE SKILL PREPARES, I APPROVE.]
- Tools or connectors it may use: [DRIVE, GMAIL, A SHEET, HIGGSFIELD, OR "NONE".]

HOW I WANT TO WORK
1. Ask me clarifying questions first if anything above is thin or ambiguous. One
   question at a time.
2. Then draft the SKILL.md. Keep the body lean. Use imperative voice. Explain the
   why behind each instruction so it generalizes. Push deep detail into reference
   files. If you find yourself writing the same helper twice, make it a script the
   skill calls.
3. Give me 2 to 3 realistic test prompts, phrased the messy way I would actually
   type them, so we can see the skill run.
4. Then optimize the description with me using the block below.

DESCRIPTION OPTIMIZATION
Write the description as the primary trigger. Make it a little pushy, since skills
tend to under-fire. Put all of the when-to-use information in the description, not
the body. Use this skeleton:
  "[ONE LINE ON WHAT THE SKILL DOES]. Use this whenever [TRIGGER SITUATIONS,
  KEYWORDS, AND PHRASINGS], even if [THE OBVIOUS CASE WHERE I WON'T NAME IT
  DIRECTLY]. Do not use for [THE NEAR-MISS CASES THAT SHOULD NOT TRIGGER]."
Then pressure-test it. Generate about 20 realistic eval queries in my world
([YOUR CONTEXT: YOUR NICHE, YOUR TOOLS, YOUR TYPICAL PHRASING]):
- 8 to 10 SHOULD-TRIGGER: vary the phrasing, mix formal and casual, include cases
  where I never name the skill, and a couple of edge cases.
- 8 to 10 SHOULD-NOT-TRIGGER: make these near-misses on purpose. They share keywords
  or topic with the skill but actually need something else. Avoid obvious throwaways.
Show me the query set to approve or edit before you tune anything. Then refine the
description so it catches the should-trigger cases and stays quiet on the near-misses.
Pick the version that holds up on the cases we did not tune against, not the one that
just fits the list. Show me the before and after.
Live, leave the slides

Build one skill,
start to finish.

Pick one simple step and build its skill in plain language. Then open the skill that builds skills.

The next layer of intent

Ask of each step: what context does it need, and what tools does it touch? When a step has to reach a live system, that is a connector.

The last layer

Predict when you'll need a step, then schedule it. Open Cowork, create a scheduled task, and the work runs without you. We go deep on this May 26.

What we covered
01

You moved to the edge. The intelligence layer changed the job.

02

AI Intuition. Where AI belongs, step by step.

03

The stack. Chat, projects, skills, artifacts, connectors.

04

A working skill, built live.

Homework, before May 26

Label every step of your workflow
with AI Intuition.

Next session
05.26
We build together.

Increase your output.
Keep your taste.

With gratitude.

44 / 44