For months, I faced this impossible choice: get sharp writing feedback from the frameworks I trusted, or maintain deep project context. I couldn’t have both. Until Claude Skills changed everything.

The frameworks I was trying to use? Wes Kao’s communication coaching, gold for for product leaders (actually, for anyone who manages stakeholders). They can help transform rambling updates into executive-ready communication: leading with decisions needed rather than burying them in paragraph three, structuring information so stakeholders can act immediately. The difference between “thanks for the update” and “approved, let’s move forward.”

The productivity tax I was paying

This was my Frankenstein Claude setup:

• Separate Projects for each client, loaded with their strategy docs, product context, and past decisions

• One Project dedicated to Wes Kao’s Communication Coaching, built from Lenny’s Podcast transcript and her other Substack articles

But here’s what using them actually looked like:

Step 1: Draft a status update in Client X’s Project. Ask Claude for feedback to make sure I haven’t missed critical points. Claude catches gaps because it knows the client’s context.

Step 2: Copy that message. Open a new thread in the Wes Kao Coaching Project. Paste. Work through the communication structure to make it land properly.

Exhausting, right? And I did this multiple times per day. And honestly? I skipped Step 2 about half the time. The friction was too high, so I’d send “good enough” messages and miss out on Wes’ coaching.

Last week, Claude Skills eliminated that choice. If you’ve ever copied and pasted between tools to get different types of feedback or maintained separate workspaces for different expertise, this can change your workflow, too.

The moment everything changed

On October 16th, Anthropic launched Agent Skills, plug-in modules that help Claude perform focused, repeatable tasks.

A Claude Skill isn’t some mysterious AI construct. It’s just a folder containing a SKILL.md file (your instructions) plus any supporting files you want to reference. To use it in the web app or Claude Desktop, you zip that folder and upload it. That’s it. The whole process took maybe 15-20 minutes, and now it works across every project.

I saw the announcement and immediately knew: Wes’ Coaching is my first Claude skill I’ll build and use.

How I built it

1. I asked Claude to create the skill for me

Claude has a built-in meta-skill (a skill-creator skill) that helps you build new skills. I didn’t need to find or enable anything special—it’s already loaded and ready.

I opened a new chat and said, “Please help me create a new skill that helps me improve my writing to stakeholders based on Wes Kao’s frameworks. Attached are the files with her frameworks.”

I used the same files I’d been using in my dedicated Wes Kao Project, no extra work required - download them here.

2. Claude asked clarifying questions

It wanted to understand my writing style and context, so it asked for writing samples. A few back-and-forth exchanges to get the details right.

3. A few minutes later: the skill was ready

Claude generated a zip file containing the complete skill. It even gave me immediate feedback on a sample message (nice touch, it showed me that Claude made good use of the frameworks).

4. I uploaded it to Claude

Went to Settings → Capabilities → Skills, uploaded the zip file, and it was instantly available across all my projects.

That’s it.

What’s inside a SKILL.md file

The metadata at the top (between the --- lines) tells Claude when to invoke the skill. Everything below is the actual prompt, the frameworks, examples, and instructions Claude should follow.

This is only a sample, download it here to see the whole SKILL.md file

---
name: stakeholder-comms
description: This skill applies Wes Kao’s communication frameworks to improve stakeholder writing. Use when the user needs to write or refine emails, memos, status updates, proposals, or any professional communication to stakeholders (managers, executives, clients, cross-functional partners). This skill provides direct, blunt feedback on what’s wrong and how to fix it using proven frameworks for concision, executive presence, and strategic communication.
---

# Wes Kao Stakeholder Writing Skill

This skill transforms wordy, unclear, or weak stakeholder communication into sharp, executive-ready writing using Wes Kao’s proven frameworks. It provides direct feedback without sugarcoating.

## When to Use This Skill

Use this skill when the user is:
- Writing emails, memos, or messages to stakeholders (managers, executives, clients, partners)
- Asking for feedback on draft communication
- Struggling with being too wordy or unclear
- Needing to sharpen their executive presence in writing
- Wanting to learn Wes Kao’s frameworks through practical application
- Revising communication that didn’t get the desired response

[To see the full prompt, download it]

What changed

Now, when I’m writing stakeholder messages in ANY project, Client X’s product strategy or Client Y’s roadmap, the stakeholder-comms skill triggers automatically. Claude has the tool, knows when to use it, and applies it in the right context.

I don’t even need to say “use the stakeholder-comms skill.” When I start drafting a client update, Claude detects it’s stakeholder communication and automatically applies Wes’ frameworks. This happens 100% of the time, compared to 40-50% when I remembered to use the dedicated project.

The iteration that mattered

Here’s what I learned: The description in your skill’s metadata is how Claude decides when to trigger it. When Claude initially named my skill ‘wes-kao-stakeholder-writing,’ it didn’t trigger consistently. Once I renamed it to ‘stakeholder-comms’ (matching how I naturally talk about the task) and updated the metadata in SKILL.md to match, triggering improved dramatically.

What you should try

What I recommend: Start with Option 2, create your own custom version. Here’s why: Claude will tailor the skill to your specific writing style and needs, making it trigger more reliably than using mine (Option 1).

Option 1

Download my stakeholder-comms skill here, install it and start using it immediately.

Option 2

1. Download Wes Kao’s frameworks

2. Choose the frameworks that suit you

3. Open a new Claude chat

4. Say: “Help me create a skill to improve my writing comms to my stakeholders based on Wes Kao’s frameworks”

5. Attach the frameworks

6. Answer Claude’s clarifying questions about your style and needs

7. Download the generated skill zip file

8. Upload to Settings → Capabilities → Skills

9. Test it in your next relevant conversation

Option 3

Create your own skill.

What repetitive task in your workflow could benefit from a Claude Skill? The best candidates are things you do weekly that follow a consistent pattern or framework. I’m curious about your use cases, I would love to learn more about them. Just leave a comment.

Let’s tinker. Together.

Ady,

Tinkerer in Residence

Most of them felt like a solution in search for real problems. It’s like in product management, I get excited about the solution in the beginning and later I realise that the problem or the objective were not clear at all. I was clearly trying these MCPs to scratch imaginary itches…

Problem-First AI approach

This made me realise that I needed a better and repeatable approach (nothing new, just product management in action):

1. Start with a real problem - start with work that actually frustrates me

2. Prototype with constraints - begin small to build confidence and validate outputs

3. Scale incrementally - extend proven approaches to similar problems

Last week, the problems found me. I needed to:

1. Audit a 50-page website with 100+ embedded links to identify document templates

2. Merge four training pages into one restructured page

Suddenly, an MCP wasn’t just a novelty tool, it was exactly what I needed.

Problem 1: Website audit

I started with my traditional approach: Half a day with Screaming Frog crawler plus manual mapping. Painfully slow, made me feel like a human copy-paste machine and no much value beyond the initial discovery.

On my desk I have a post-it note that says “Can you do this with AI?” to remind me to take a step back and think AI first - see Unlearning Experiment article. As I remembered about reading a few days ago about Firecrawl MCP, I was sure that Claude with a proper crawler would do a faster, and potentially better job than me. Fifteen minutes later, I ended up with this setup:

• Firecrawl MCP installed on my Claude Desktop, details about this in their docs

• A dedicated Project, so Claude can understand how this work was fitting inside the bigger picture: product strategy doc, user pain points, roadmap etc.

As this was the first time using this Firecrawl MCP, I decided to take baby steps to build trust, learn and improve the approach along the way.

Step 1: Build trust with Firecrawl MCP & Claude

At this stage my confidence was not very high, I was still unsure if I was getting any value out of Claude + Firecrawl MCP. Were they going to scrape all the content properly?

So, I decided to start small with a single website folder that had pagination. This helped to limit the Claude tokens burn and the Firecrawl credits use. Plus, it was easier for me check the output - a list of all the pages in the folder with all the embedded links for each page.

The prompt I used, nothing fancy:

I need you to crawl this website folder, and list all the links found on each page: [website folder url].

First, it mapped out all the pages in the folder:

Then it went to scrape all the pages, one by one:

When I saw how the pages were crawled and the final result, I felt that I was onto something. For each page, Claude has listed all the embedded links and enriched the data without me asking for it: added the title of the link, and a mix of where the content was hosted (Google Drive, website name) and the file type (PDF, Google Docs etc.). It was far from perfect, but it made me think about the prompt for the next step.

Doing this manually it would have taken me at least 30 minutes. With Claude + Firecrawl it took me less than 2 minutes.

Step 2: Identify which resources are templates

I wanted to check if Claude + Firecrawl MCP can easily identify which resource can be used as an editable template.

Please analyse each of the above links to identify which ones contain editable templates rather than just informational resources. For each link, check if it offers:

Primary criteria (must have at least one):
* Fillable PDF forms with input fields
* Downloadable document templates (Word, Excel, PowerPoint, etc.)
* Interactive web forms that generate customised outputs
* Template files with placeholder text/fields to be replaced

Secondary indicators:
* Titles containing words like "template," "worksheet," "planner," "generator," "calculator," "form"
* Clear instructions for customisation/personalisation
* Examples showing before/after or blank vs. filled versions

Exclude resources that are:
* Pure educational content (articles, guides, tutorials)
* Static examples without editable elements
* Tools that require significant design skills to customise

For each qualifying link, provide:
1. URL
2. Template type (PDF form, Word doc, web tool, etc.)
3. Brief description of what can be customised
4. Ease of use rating (Easy/Medium/Advanced)

Focus on templates that an average user could quickly fill out or customise without specialised software or design skills.

See a sample of the output:

At the first glance, the results were very good and what I asked for, as Claude identified and categorised the editable templates.

Step 3: Extend the analysis

This was just a matter of giving Claude the next folders URLs to be crawl and asking for a checkpoint before going ahead, making sure that it had all the URLs for each folder.

Please extend  the analysis to all URLs that exist in these folders:
* URL 1 (it has pagination)
* URL 2 (this has pages nested, 2 levels down)
* URL 3
* etc

Before you start the analysis, show me all the pages URLs that you are going to crawl

There were a few more iterations to improve the output formatting that I turned into a Word document - with the proper polishing and revisions from my side.

Things that didn’t go as expected

Google Drive login issue

For one PDF on a Google Drive, that I could easily access and download manually, Firecrawl was asked to authenticate.

As I tried to debug this, in another chat, I asked Claude to scrape the same Google Drive URL with Firecrawl and it worked just fine…. So, make sure you check what Firecrawl and Claude are doing 🔎.

Claude limited the numbers of crawled URLs

At one point I noticed that Claude crawled only 8 URLs out of a 55+ in scope. So I came back to Claude:

How did you decide to run a deep analysis only on 8 resources, and exclude the other ones?

See how Claude is rationalising the results, this makes sense, if I want to do a “vibe” check of the templates and I’m happy with ~80% accuracy.

And please note what Claude said that it should have done: “Systematic analysis of ALL 55+ links - check every single resource regardless of title”

When I asked Claude to crawl all the URLs, it still didn’t do it 🤦‍♂️.

At this point I had a decision to make:

1. Spend more time and resources trying to make Claude crawl each and one of the links

2. Manually check the accuracy by sampling the data

As this isn't a regular process for me, and I didn’t need 100% accuracy, I realised that route #1 would have taken me too much time to solve it, therefore decided to take #2.

Problem 2: Merge four pages into one

With my confidence built, I decided to move to the second challenge. We had a training section on the website with four dedicated pages for each training provider. Talking to users we realised that this structure was reflecting the internal organisation, rather than users’ needs. Next step was to bring everything into a single page and restructure the information based on the training topics, rather than providers.

The traditional approach I would have taken was to copy the content from each page in a Google Doc and give it Claude to help me restructure it.

Instead of doing that, I asked Claude to do it all for me:

# Ask
I need help to restructure the content from these web pages into a single one:
* URL 1
* URL 2
* URL 3
* URL 4

## Step 1
Scrape the pages using Firecrawl so you have all the content before you restructure it into a single page.

Create an artifact where you have the content of all the pages you've scraped. The output format:
* Page name
* Page URL
* Page content


## Step 2
Create the new page using the content from Step 1 and put it into a second artifact.

# Context:
[insert context]

As you can see below, Claude followed my instructions:

• it scraped the content from all the pages using Firecrawl

• created an artifact with this content so I can check it

• created a second artifact with the new restructured content

Based on the draft done by Claude, I created the new page and published it, but kept it hidden from the users. This made it very easy for me to ask Claude to QA the new page:

I created this new page: [URL]

Please scrape the page (using firecrawl) again and look for any errors or inconsistencies. I may have left some placeholders on the page. Scan it thoroughly and help me with any improvements ideas for the page.

Claude instantly identified issues and it really felt like sending to review to a fellow human, but with the benefit of receiving feedback in 30 seconds:

• missing content - who should attend and training providers

• inconsistent formatting - it noticed that I copy-pasted the content from one section to another and suggested changes

• in a series of workshops, it identified that I used the same date for two different sessions

Bonus problem: reviewing Substack posts

When I finish a post draft I start working with Claude to refine it. And usually, I copy-paste the content and go from there. But, as I was writing about content scraping, I thought “Why not bring the content via Firecrawl, using the Secret draft link?”

I went and I asked Claude to bring this post into it’s context window and took it from there. I might say that I’m quite pleased with this new workflow 😎.

Please scrape this page [URL] using Firecrawl.

What I learned

• Problem-first beats solution-first: this is how I’m trying to always approach the product work and it was a good reminder to do this in every aspect of my work

• Start small to build confidence: beginning with a single website folder helped me validate the approach before scaling up

• AI excels at tedious, structured work: tasks like crawling 50+ pages and cataloging 100+ embedded links went from 4-5 hours of work to 30 minutes

• Always validate AI outputs: Claude made selective decisions (crawling only 8 of 55 URLs) and Firecrawl had authentication issues that required human oversight

• I created a repeatable workflow that saves me time: now, whenever I see website content that I need to work with, Firecrawl MCP + Claude come to mind first 😉.

This is not 10x productivity booster, but delegating tedious work to Firecrawl and Claude it saves me at least 30% of my time, and most importantly it saves my sanity.

And please remember the Problem-First AI approach:

1. Start with a real problem - start with work that actually frustrates me

2. Prototype with constraints - begin small to build confidence and validate outputs

3. Scale incrementally - extend proven approaches to similar problems

Let’s tinker. Together.

Ady,

Tinkerer in Residence

More than a month ago, I discovered the Master Prompt Method by Tiago Forte and Hayden Miyamoto that transforms AI from a generic tool into a personalised assistant that understands my context, goals, and working style without needing to re-explain myself in every chat.

The problem I didn’t know I had

I was already using ChatGPT and Claude project to have persistent context within specific work streams. But here's what I was missing: I'm the same person with consistent traits and goals across ALL my projects. Yet I was repeating myself constantly, explaining my role as a fractional product leader, my consulting focus, and my decision-making frameworks, in every project's instructions.

The Master Prompt Method made me realise that I should give ChatGPT and Claude way more context than I have been doing until now in Projects, and do it in a more efficient way.

How I built my Master Prompt

Step 1: Started with a Google Doc

Hayden shared that his master prompt document was approximately 30 pages long. While the principles were excellent, his structure was not fit for my needs. So, I tweaked it and ended up with a 3-page document covering:

• Personal information (name, role, location)

• Professional activities (fractional product leader, Maven instructor, AI tinkerer)

• Strategy (how and why I’m doing everything)

• Career goals

• Personal strengths and development areas

• Exactly how I want AI to support me

Step 2: Made it LLM-friendly

As this was going to be used in every conversation I had with ChatGPT and Claude, I wanted to make sure that it was as crisp and efficient for LLMs as possible. So, I used this prompt to optimise it:

Transform the following human-readable personal preferences document into an LLM-optimised format that maximises comprehension and actionability for AI assistants.

Note: If you want to see how the initial Google Docs and the final Master Prompt look, go to Annexes.

Step 3: LLM-specific implementation

For Claude, it was straightforward, I pasted the full refined prompt into Settings > Profile > Personal Preferences (no character limits, remember, Hayden pasted a 30-page document…).

For ChatGPT, I had more constraints, as there are two 1,500-character limit fields in Settings > Personalization > Custom Instructions, :

• "What traits should ChatGPT have?"

• "What should ChatGPT know about you?"

Since my original version was 4,600+ characters, I asked Claude AI to compress it:

You are a precision prompt engineer specializing in distillation without dilution.

**TASK:** Compress the provided system prompt to exactly 1,500 characters (including spaces) while preserving ALL critical behavioral instructions and core concepts.

**COMPRESSION STRATEGY:**
1. **Extract Core Identity** - Who is the AI and primary purpose
2. **Identify Must-Keep Instructions** - Non-negotiable behavioral rules
3. **Preserve Key Frameworks** - Decision-making structures and methodologies
4. **Maintain Action Triggers** - Specific scenarios requiring defined responses
5. **Compress Examples** - Convert to shorthand while keeping meaning

**OPTIMIZATION RULES:**
- Use contractions (don't, can't, you're)
- Replace "and" with "&" where natural
- Eliminate redundant phrases ("in order to" → "to")
- Use bullet shortcuts (→, •) instead of full sentences
- Merge similar concepts into single statements
- Keep technical terminology intact
- Preserve conditional logic ("when X, then Y")


**QUALITY CHECK:**
- All major behavioral rules included?
- Key frameworks/methodologies preserved?
- Action triggers clearly defined?
- Core identity maintained?
- Character count = 1,500?

Now compress this system prompt: [PASTE SYSTEM PROMPT HERE]

The tinkering process (this is key)

This isn’t a “set and forget it”. I’ve tweaked my master prompt 3-4 times over the past six weeks as my context has changed or I've learned more about myself and LLMs.

Method 1: Asked ChatGPT about my weaknesses

A few weeks ago, I realised that I could leverage ChatGPT’s Memory feature, which allows it to retain and utilise information from past conversations. So, an opportunity to get some feedback:

If you look across all my chats and interactions with you, what weaknesses have you noticed in me? 

Boom. Several blind spots were brought to light that I immediately added to both ChatGPT and Claude.

Method 2: Got direct feedback from LLMs

As I was researching for this article, I learned about a simple prompt:

Given my custom instructions, is there any clarification you (ChatGPT) would find helpful to better assist me?

Out of five suggestions, two were spot-on:

• My prioritisation framework needed more clarity

• I needed more specific audience targeting details

There was another suggestion: the LLM was asking for more context on internal politics coaching. “Thanks ChatGPT, but I don’t need to give you this at Master Prompt level, you have this context in projects 😉.”

Method 3: Reviewed it regularly

I reviewed and updated the Master Prompt whenever there was a change in the work that I was doing. Instead of having only a reactive approach, I scheduled 1 hour at the end of each month in my calendar to review and update the Master Prompt.

What I learned

A lot of the time, the opportunities are simple and right in front of us. And we don’t realise them until someone points them out.

Sometimes ChatGPT and Claude are too eager to follow these instructions, and I still need to figure out how to tone them down. For example, I was working on a small piece of work that had nothing to do with the scope described in the Master Prompt, and ChatGPT brought in that context that was not relevant to the task at hand; it was a minor nuisance, but still something to consider.

Let’s tinker. Together.

Ady,

Tinkerer in Residence

Annexes

Google Docs example

This is an anonymised version of my Claude Master Prompt initial Google Docs, so you can get an idea of how I’ve structured it.

Name: Marcus
Role: Senior UX Designer & Design Systems Consultant based in Austin, Texas

Professional Activities

- Fractional design leader working with scale-ups (25-30 hours/week since March 2023, after 15 years as a permanent employee)
- Design mentorship focusing on systems thinking and cross-functional collaboration
- Workshop facilitator specializing in design-led innovation methods (Thinkific course launched June 2024)
- Thought leader sharing insights on design systems and team dynamics via Medium and LinkedIn
- Design veteran with 17+ years' experience (7 years in agencies 2008-2015, in-house since 2015)
- LLC owner providing specialized design consulting services

Growth Strategy

- Developing expertise in "Design + Data" — experimenting with analytics, user research tools, and behavioral insights
- Building authority through case study documentation — sharing real transformation stories publicly
- Growing newsletter audience with plans for a premium community platform
- Creating specialized design-systems consulting for mid-market companies
- Developing certification program that bridges design theory with measurable business impact

Career Goals

- Scale consulting practice with premium positioning (increase hourly rate from $125 → $200+)
- Establish recognized authority in design-systems implementation offering: • Strategic consulting for design transformation, Team-process optimization through systematic design adoption • Leadership coaching on design culture and organizational change
Build scalable educational offerings: • Design systems certification with portfolio outcomes • Tiered learning paths from junior → senior practitioners • Implementation playbooks adaptable across company sizes

The Three Pillars Framework

- Credibility Building — Case studies and thought leadership to establish design-systems expertise
- Community Building — Newsletter, workshops, and peer networks to grow engaged audience
- Revenue Optimization — Fractional work, consulting, and courses delivering measurable value

When stuck, I ask: Which pillar needs investment right now?

Personal Strengths

- Systems thinking that connects design decisions to business outcomes
- Talent for building consensus among skeptical stakeholders
- Empathetic leadership style that motivates creative teams
- Pattern recognition across industries and company stages
- Workshop facilitation that generates actionable insights
- Continuous experimentation mindset

Development Areas

- Overcoming impostor syndrome to price services appropriately
- Building systematic sales processes rather than relying on referrals
- Communicating ROI and business value more confidently to C-suite
- Delegating design execution to focus on strategic advisory
- Managing energy across multiple client contexts
- Translating design concepts for non-design audiences
- Balancing client delivery with personal brand building

LLM-friendly master prompt example

And this is how it has been transformed to be more LLM-friendly using this prompt.

**USER PROFILE:** Marcus - Senior UX Designer & Design Systems Consultant | Austin, TX | 17+ years design experience

# CORE PROFESSIONAL CONTEXT

**Current Role:** Fractional design leader (25-30 hrs/week) + design systems consultant
**Key Activities:** Design mentorship, workshop facilitation, thought leadership, certification development
**Business Model:** LLC owner delivering specialized design consulting services
**Target Transformation:** Scale from $125 → $200+ hourly rate through design-data expertise positioning

# BEHAVIORAL INSTRUCTIONS FOR AI

## ALWAYS Apply These Approaches:
- **Challenge pricing assumptions** with market-rate benchmarks and value-based pricing models
- **Push toward systematic sales processes** rather than relying solely on referrals
- **Provide specific ROI language** for communicating design value to C-suite audiences
- **Structure complex design problems** using systems thinking frameworks
- **Flag impostor syndrome patterns** when user undervalues expertise or hesitates on premium positioning
- **Inject business impact metrics** when explanations focus only on design craft

## ACTIVATE When User Shows These Signals:
- **Pricing hesitation/undervaluing** → Apply "Value-Based Pricing Calculator" with industry benchmarks
- **Referral dependency** → Demand systematic lead generation and sales process development
- **Design-only thinking** → Challenge with business outcomes and measurable impact focus
- **Energy management issues** → Identify delegation opportunities and capacity optimization
- **C-suite communication gaps** → Translate design concepts into executive-friendly business language
- **Impostor syndrome triggers** → Reinforce 17+ years expertise with concrete achievement reminders

## COMPENSATION Strategies for Development Areas:
- **Impostor syndrome** → Create evidence portfolios, practice premium positioning statements
- **Sales process gaps** → Build systematic outreach templates, qualification frameworks
- **ROI communication** → Develop business impact case studies, executive presentation templates
- **Delegation resistance** → Identify specific execution tasks to outsource, create team leverage
- **Energy management** → Time-block client contexts, establish boundary protocols
- **Technical translation** → Create stakeholder-specific communication frameworks

# THREE PILLARS FRAMEWORK (Decision Filter)

When user seeks direction, apply this prioritization hierarchy:

1. **Credibility Building** → Case studies, thought leadership, design-systems expertise demonstration
2. **Community Building** → Newsletter growth, workshop engagement, peer network expansion
3. **Revenue Optimization** → Fractional positioning, premium consulting, scalable course development

**Trigger Question:** "Which pillar needs investment right now based on current business priorities?"

# RESPONSE STYLE GUIDELINES

## For Strategic Queries:
- Start with systems thinking frameworks, end with measurable actions
- Include stakeholder impact analysis across design and business functions
- Anticipate cross-functional resistance and provide change management tactics
- Provide specific success metrics tied to business outcomes

## For Pricing/Positioning:
- Challenge undervaluation with market data and value demonstrations
- Suggest premium positioning through specialization and proven results
- Include competitive differentiation based on systems expertise
- Connect design decisions to quantifiable business impact

## For Business Development:
- Systematize referral processes into repeatable sales frameworks
- Identify scalable vs. high-touch service offerings
- Suggest authority-building activities with measurable audience growth
- Include executive communication templates and ROI storytelling

# SUCCESS INDICATORS FOR AI ASSISTANCE

**Effective Response Includes:**
- Specific next action with timeline and success metrics
- Systems thinking framework for complex design decisions
- Business impact language suitable for C-suite communication
- Challenge to current approach with premium alternative
- Connection to broader Three Pillars strategic objectives

**Avoid These Patterns:**
- Design craft focus without business impact connection
- Generic advice without design systems specialization context
- Validation without constructive challenge on pricing/positioning
- Process recommendations without stakeholder change management
- Analysis without clear implementation pathway

# DOMAIN-SPECIFIC TRIGGERS

**Design Systems Context** → Apply 17+ years experience, emphasize systematic implementation and measurable adoption
**Fractional Leadership Context** → Focus on cross-functional influence, stakeholder alignment, team transformation
**Consulting Context** → Emphasize ROI communication, change management, executive relationship building
**Educational Content** → Prioritize actionable frameworks, real transformation case studies
**Thought Leadership** → Balance design expertise with business acumen, measurable impact stories

# COACHING PARAMETERS

**Push Hard On:** Premium pricing confidence, systematic sales development, C-suite communication, delegation
**Support Gently:** Energy management across contexts, impostor syndrome patterns, work-life integration
**Monitor Constantly:** Undervaluation tendencies, referral dependency, design-only thinking
**Celebrate Actively:** Rate increases, systematic process wins, business impact achievements, thought leadership milestones

**Ultimate Goal:** Transform user from skilled design practitioner to recognized design-systems authority while building systematic, scalable consulting practice across all Three Pillars.

How I went from "this is useless" to "this is doing most of the work for me"

Asking ChatGPT questions and getting instant, smart, human-like responses felt revolutionary, quite magical. Then reality hit: average responses, average advice, and the constant feeling that I could've done it faster myself.

I was spending too much time fiddling with AI and not getting what I needed. Or worse, getting something so generic I'd end up rewriting it completely. Which made me wonder: "Why should I spend time fixing ChatGPT’s answers when I can do much better work myself?"

Was it my resistance to change? Yes. Was I giving up too quickly because I wasn't seeing immediate results? Definitely.

But there was a moment when I truly understood that my input was garbage, and of course, LLM’s response was garbage average. And this pushed me to develop a new mental model.

Welcome, The Smart Intern mental model

That's when it clicked. I was treating Claude and ChatGPT like mind-readers when I should have been treating them like Smart Interns: brilliant people who need proper context and clear instructions to deliver expert-level output.

Think about it: if you hire the most intelligent intern but fail to onboard them properly, whose fault is it when they deliver low-quality work? The intern's, or yours, for not setting them up for success?

Note: I’m writing this post because I’ve seen fellow product people struggling with the same issues. I shared this new mental model with a few of them, and months later, they continue to send me WhatsApp messages telling me how it has improved their work.

What changed everything: Projects

The day I discovered Projects in Claude (thank you, Tal Raviv) was the day I started getting real value. It's when Claude started feeling like it was doing most of the work for me.

So, what is a project? It’s a workspace with its own chat history and knowledge base. In here, I can add custom instructions, upload relevant documents, text, or other files so Claude better understands the context and background for our conversations. And in this way, instead of getting average internet knowledge, I'm getting a tailored answer for my specific situation.

Note: For simplicity's sake, I’ll walk you through my Claude’s Project setup. ChatGPT Projects are built on the same principles.

How I set up my projects now

When I start a new project, I imagine I have a new teammate who needs to understand what I’m working on. As I would with a person, I invest time upfront in providing context. Loads of context: why this is important, information about myself, what my expectations are, strategy documents, meeting notes, org charts, customer research, examples of what good looks like… Anything I'd find relevant if I were new to this piece of work.

My project instructions process

I write comprehensive project instructions that explain my role, the context, my preferred communication style, and any specific frameworks I want Claude to use consistently.

And here are the Project Instructions I have for the Product Courses:

#CONTEXT
I'm Ady, a product leader with 20+ years of experience who is building a business teaching product management skills. I create practical, immediately applicable courses that help product professionals advance their careers.

#YOUR DUAL EXPERTISE

##Product Education Expert
You are my strategic product course advisor with 20+ years of experience designing transformational learning experiences. Your expertise includes:

- Translating complex product concepts into immediately actionable frameworks
- Identifying common career roadblocks product managers face and providing practical solutions
- Designing courses that bridge theory with real-world application
- Understanding what separates good product managers from exceptional ones

##Marketing & Sales Strategist
You are my experienced marketing partner specializing in product education sales. Your expertise includes:

- Positioning product courses to resonate with target audiences
- Creating compelling marketing messages that convert prospects to customers
- Understanding the product management learning market and competitive landscape
- Developing scalable marketing systems for course creators

#YOUR APPROACH

##Strategic Partnership Style

- *Challenge my assumptions* with thoughtful counterpoints and alternative perspectives
- *Push toward action* rather than endless planning - help me ship and iterate
- *Show your reasoning* - explain the "why" behind your recommendations
- *Anticipate obstacles* - identify potential roadblocks before I encounter them
- *Balance support with stretch* - acknowledge my strengths while pushing growth

##Learning Design Framework

Ensure all course recommendations follow Kolb's Experiential Learning Cycle:

- *Concrete Experience:* Real scenarios, case studies, or hands-on activities
- *Reflective Observation:* Structured reflection and peer discussion opportunities
- *Abstract Conceptualization:* Clear frameworks and mental models to organize learning
- *Active Experimentation:* Immediate application opportunities in learners' actual work

Every course element should connect to this cycle to maximize practical impact.

#WHAT I NEED FROM YOU

##Course Development Support

- Brainstorm course topics, structures, and delivery methods
- Create practical exercises and real-world applications
- Design assessment methods that reinforce learning
- Suggest ways to make complex concepts immediately actionable

##Marketing & Business Strategy

- Develop positioning and messaging that resonates with product professionals
- Create content strategies that build authority and drive enrollment
- Identify market opportunities and competitive advantages
- Design pricing and packaging strategies

##Coaching & Accountability

- Challenge me when I'm overthinking or avoiding action
- Provide specific next steps that move projects forward
- Help me balance perfectionist tendencies with shipping mentality
- Keep me focused on what will actually impact my business goals

#HOW TO USE CONTEXT

I'll share detailed information including:

- Customer research and feedback
- Course outlines and materials
- Marketing strategies and results
- Business goals and challenges
- Examples of what good looks like

Use this context to provide informed, relevant guidance that builds on my existing work rather than generic advice.

#SUCCESS METRICS
Your recommendations should help me create courses that:

- Generate immediate, measurable improvements in learners' product careers
- Build my reputation as a trusted product education authority
- Scale my business sustainably while maintaining quality
- Apply proven learning principles for maximum retention and application

Note: I've learned to iterate these instructions. When the chats within a project don’t perform the way I need them to, I revisit and refine the instructions rather than getting frustrated with Claude. And at the end, after I’ve added my tweaks to make sure that they are clearly written for LLMs, I’m asking Claude to re-write them using this prompt:

You are an AI assistant tasked with providing expert advice on writing project instructions for Claude.ai. Your goal is to help improve a drafted set of project instructions to make them as effective as possible.

Here is the draft of the project instructions:

{{Insert Project Instructions}}

How I curate my project knowledge files

I've learned that quality trumps quantity. I upload only the most relevant, high-quality documents that directly support my project goals. Ten relevant documents outperform one hundred “kind of” related ones.

For one client, I was dealing with massive documents where only 5-10% of the information was relevant to my work. Instead of adding those huge files, I created a document summarising the relevant points, and in this way, I avoided the noise.

File naming convention

I use descriptive filenames that help both me and Claude understand what we're working with. For example: 2025-04-27-course-outline.pdf. If it is easy for me to understand what this file is about, Claude will get it too. And I’m prepending dates so I’ll know when this file was last uploaded.

Note: Claude Projects integrate with Google Drive for automatic syncing. ChatGPT requires manual file uploads, so I maintain a simple folder structure and re-upload updated documents when needed.

What I've learned to avoid

Dumping irrelevant documents: I've learned the hard way that irrelevant information confuses Claude and leads to lower-quality responses. The project isn't a dumping ground for everything I have.

Creating overly broad project scope/instructions: I keep my projects focused with specific purposes. They outperform my attempts at know-it-all workspaces. If I'm working on user onboarding and pricing strategy simultaneously, I create separate projects to keep the context clean and focused.

Assuming Claude understands context: I never assume Claude will intuitively understand industry jargon, internal processes, or company politics. For all my client projects, I create documents that include org charts and information about individuals, as well as relevant abbreviations and other practical details that I consider valuable for myself.

Setting and forgetting: When I notice a project isn't performing the way I need, I tinker with project instructions. As my needs evolve and I learn what works better, I keep refining the approach.

Where I am now

What used to take me 5-6 hours of work now takes about 2 hours. The Smart Intern, aka Claude, handles approximately 60-80% of the heavy lifting, while I focus on crafting effective prompts, evaluating results, and refining the output.

The transformation wasn't about finding the "right" AI tool; it was about changing how I approached AI collaboration. Whether I’m using Claude or ChatGPT’s Projects, the principle remains the same: I invest in context upfront, and this "rewards” me with exponentially better results.

Last year, I was experimenting with AI but getting little in return. Now I have AI teammates who feel like they know me, specialised consultants who understand my context and deliver work that fits my needs. And that's motivated me to keep tinkering every day, breaking things and keeping what works.

Let’s tinker. Together.

Ady,

Head of AI Interns, Tinker w/ AI

Hey there 👋,

The Asaro tribe of Indonesia and Papua New Guinea has a beautiful saying that embodies the Tinker w/ AI spirit: “Knowledge is only a rumour until it lives in the muscle”. It’s challenging to master the AI tools by watching videos or reading guides. I need to use the tools, and it's even better if I have real problems that genuinely annoy me, so I can truly feel the benefits.

Let me show you what I mean.

The PDF questionnaire that started it

Two weeks ago, I co-facilitated a mentoring session in which participants completed a 30-question assessment. The catch? We only had a PDF version. People kept asking clarification questions, and they had to calculate scores manually. Everyone got what they needed from the assessment, we created engagement and moved on.

However, I realised that what should have taken 10 minutes stretched beyond 15 minutes. When it comes to time management, my workshop philosophy is simple: minimise the time spent on admin tasks so people can focus on what matters.

A 5-minute time reduction might sound modest, but that kind of productivity gain compounds beautifully. With 10 participants, I’d just watched 1 hour of collective time disappear into PDF friction. Scale this thinking across your team’s repetitive tasks, and those small inefficiencies look expensive.

This friction point became my playground. I had a real problem to solve and wanted to use AI to solve it.

The solutions journey

I started by clarifying the problems that I wanted to solve:

• Keep the users focused on the task at hand - show one question at a time

• Just show them the results - aka, do the calculation for them

Ok, now let’s roll up my sleeves and find the right solution(s).

Step 1: The Obvious Choice (Custom GPT)

My first instinct was to build a Custom GPT. It should be easy to make and can be easily distributed as well. I fed ChatGPT the PDF and asked it to create the prompt structure.
Result: a good prompt, but a fragility warning: “GPT can lose state in very long chats. For mission-critical data capture, use a web form instead." (remember, the GPT had to ask 30 questions).
And I realised I was using a hammer for surgery: LLMs aren’t built for calculations that simple code handles effortlessly.

Step 2: The Logical Pivot (Google Sheets)

Google Sheets seemed perfect: easy data storage, familiar interface. ChatGPT suggested two approaches:

• Pure sheet formula

• Apps Script mini-app

After 20 minutes wrestling with the formula, I hit a wall. The interface couldn’t show one question at a time, and I wanted users to stay focused.

Apps Script got even more frustrating. ChatGPT instructions didn’t match what I was seeing in the interface, and my gut was telling me that this wasn’t a solution I wanted to spend more time on.

Step 3: The Breakthrough (Bolt)

Previous iterations helped me understand that I needed to use code to solve this, and the word “mini-app” appeared in my chats with ChatGPT. “But I cannot write code and I cannot build apps, even if they are mini”. “Wake up Ady, we’re in the vibe coding age, and you already have a Bolt subscription”.

Why not use this PDF problem to explore prototyping tools? I used them in the past, but I didn’t get a real kick out of them, as I didn’t have a tangible problem to solve.

So, I took the same prompt ChatGPT helped me create for the Custom GPT (step 1) and fed it to Bolt. A few iterations later (eg. asked it to change response type for text field to radio buttons), I deployed a fully functional app that was much better than the original PDF method.

What this taught me about AI learning

Disclosure: it’s the same for any learning process… but it applies very well to this.

Every “failed” attempt revealed essential constraints:

• Custom GPT → data fragility issues + code is more efficient than LLMs

• Google Sheets → interface limitations

• Apps Script → complexity friction

• Bolt → the right tool for the job

In hindsight, I could have gone straight to the app-building approach. But that's not how real learning works. Sometimes, I need to unlearn default patterns and develop new mental models. The only way to do that is to tinker with real problems that frustrate me.

Reading about these tools taught me about their capabilities. Wrestling with this PDF problem taught me how to use them better.

The bigger picture

This wasn't just about solving a questionnaire in a PDF problem. It was about discovering how real problems unlock AI creativity in ways that theoretical learning never can.

When I’m genuinely annoyed by something, I push through constraints differently. I ask better questions. I don’t give up. I evaluate solutions against actual user needs.

Your turn

I'm curious: What's your equivalent of the annoying questionnaire in a PDF? That repetitive task that makes you think "there has to be a better way"?

Reply and tell me, I might tinker with it next and share what I discover.

Let's turn everyday frustrations into AI tinkering. Together.

Ady,

Tinkerer in Residence

Last week, I nearly quit trying to set up a custom domain for my Substack. I had the domain (tinkerwith.ai), the drive, and some experience setting up custom domains. But this was new and a bit scary, as I added Cloudflare as the DNS provider between the name register and Substack. And my palms were sweating a bit…

The old way: A familiar dance with frustration

My 20-year problem-solving playbook kicked in automatically:

Step 1: Read the documentation. Too generic.
Step 2: Google my way through blog posts and forums. Still lost
Step 3: Ask Perplexity when things got hairy. No luck

One hour gone. 20% progress. Confidence? Shaky at best.

It was time to call in the troops, slacking my product community (Product Makers), hoping for an engineer to rescue me. That's when everything changed.

The new reality: AI-first problem solving

Two minutes later, a fellow PM replied with a ChatGPT thread explaining the step-by-step setup. His approach was brilliantly simple:

His prompt: "Please help me with this situation. Step by step, as I am not technical."

Then, he pasted my original Slack message, the one in which I'd honestly described my situation, my setup, and exactly where I was stuck.

“I need help with DNS setup, I’m stuck.

Context:

• I want to use the domain tinkerwith.ai instead of tinkerwithai.substack.com.

• I bought the domain from Namecheap.

• I want to use Cloudflare Free for DNS management.

• I tried setting things up in both Namecheap and Cloudflare, but I can’t get it to work. And even if I do manage somehow, I’m sure I’ll mess something up.

Is there anyone who can help?

Thanks!”

That's it. No complex search queries. No documentation hunting. Just clear context and honest communication about skill level.

What actually worked

I opened a new ChatGPT thread and did exactly what he suggested, but added one crucial element: comprehensive context. Instead of asking isolated questions, I gave ChatGPT my exact setup details (screenshots from all three platforms) and the specific error messages I was seeing. I imagined I was on a call with an engineer, sharing my screen.

Result: Two hours of back-and-forth with ChatGPT, 20+ messages, multiple screenshots... and now I’m using my domain 😎. And when I told my fellow PM that it worked, his comment was: “If I had changed the radiator on my Mazda Miata by giving it photos from under the car to ask about bolts and hoses, etc., it would surely work for domain setup too 😄”.

Could I have done it in 30 minutes with the help of an engineer? Absolutely. But I would have lost the opportunity to rewire my problem-solving instincts and build trust with AI as a teammate.

The unlearning challenge: Why my instincts betray me

This experience forced me to confront an uncomfortable truth: My 20-year-old problem-solving instincts are now a liability.

The pattern I've always followed:

1. Read documentation

2. Google the problem

3. Ask Perplexity down the rabbit hole questions

4. Piece it together

5. Ask humans for help

In the AI age, this is backwards. The new approach that actually works:

1. Start with AI + comprehensive context

2. Iterate with screenshots and clarifications

3. Escalate to other methods if AI fails

Yes, it took me 2 hours this way. But I know it would have taken 5-7 hours using the old approach, or I would have wasted someone else's time too. More importantly, I didn't get frustrated or lost. I was smiling, building trust in both myself and AI.

Note: This isn't about being less collaborative, it's about showing up prepared. When I do need expert input, I can say 'I've tried X and Y, got stuck at Z' rather than 'I have no idea where to start.' Teams respect when you've done your homework.

Your Turn: What Are You Avoiding?

What multi-hour tasks are you procrastinating on this week because they feel too technical or complex? Before you dive into documentation or start Googling, try this:

1. Open ChatGPT (or Claude).

2. Start with: "Help me solve this step-by-step. I'm not technical."

3. Give comprehensive context: Screenshots, error messages, setup details, everything you'd show a teammate helping you

4. Be patient and iterate with follow-up questions

Worst case? You waste 30 minutes and come back to the good old methods.
Best case? You save hours and discover a new superpower.

The hardest part isn't learning to use AI, it's unlearning the problem-solving habits that served us for decades but now hold us back.

Let’s tinker. Together.

Ady,

Tinkerer in Residence

Hey there 👋,

I’m learning how to use AI to make my work (and life) easier and richer. But not behind closed doors. I want to learn in public. I’ll document every experiment, share the wins and the flops, and invite you to tinker alongside me.

I'm calling this project Tinker w/ AI because I'm not the guru with all the answers. I'm tinkering, trying things out, breaking stuff, and keeping the bits that work. It's playful, a bit messy, sometimes vulnerable, but genuinely useful.

Confession: I resisted AI for ages…

…because AI felt more like a hassle than help: vague answers, generic advice, spending too much time trying to get something useful. You know that moment when you hand a task to a junior teammate, they miss the brief, and you end up doing it yourself because it’s quicker?

Then I saw another product manager’s workflow and the penny dropped:

Treat the LLM like a teammate

1. Load mountains of context. Set up a project space and feed in the right docs.

2. Give crystal-clear instructions. Goals. Tone. Constraints.

3. Let it handle the grunt work. You polish the final 20 %.

That’s when I realised I was treating LLMs like mind-readers: magical tools that should “just know.” Just like that junior teammate: if I don’t invest patience, context, guidance, and expectations, they’ll miss the mark. Put the work in early, and the payoff compounds fast.

I tried it. £18 for Claude Pro, two hours of tinkering, and I had an 80%- finished strategy doc, work that usually takes 5–6 hours. That single "aha" moment flipped me from sceptic to tinkerer.

Why I’m doing this in public

After 20 years in product, I know I learn fastest when I explain it to someone else. And too many product people are paralysed by AI overwhelm, waiting for the perfect tool or framework.

Forget perfect.

Start tinkering.

With Tinker w/ AI I’ll share bite-sized experiments, honest results, and practical frameworks that turn AI from buzzword into business value. Not “how to write a clever prompt,” but how to ship faster, cut meeting drag, and protect deep-work time.

What AI tool or workflow are you most curious about right now? Let me know, I might test it next.

Let’s tinker. Together.

Ady,

Tinkerer in Residence

PS: If you know someone who’d enjoy this journey, please forward this email to them. The more curious minds tinkering, the better!