“Write in a professional but warm tone. Use short paragraphs. Australian English spelling. No jargon. Include an executive summary. Format the headers like this. Structure the sections like that.”

Every. Single. Time.

Claude is brilliant. But it starts every conversation knowing nothing about you. Not your business, not your standards, not the way you like things done. You’re onboarding a new hire from scratch, multiple times a day.

That’s the biggest friction point in working with AI right now. The context gap.

I co-founded Finder and bootstrapped it to $100M+ in revenue. Now I run a growth advisory and I live in Claude Code every day. Claude Skills have become a critical part of my productivity stack.

One thing keeps proving itself: a genius with no context will always lose to a good operator with great context.

Same goes for AI.

This guide shows you how to fix that. Here’s what we’ll cover:

• What Claude Skills are and why they matter

• Where Skills work across the Claude ecosystem

• The anatomy of a good Skill (with a template you can copy)

• 5 Skills every professional should build first (with full examples)

• How to build a Skill in 5 minutes, step by step

• What I’ve actually built with Skills (real examples from my setup)

• Debugging and power user tips

What are Claude Skills (and why most people are missing out)

The problem: every conversation starts from zero

Think about hiring a new team member. Even if they’re brilliant, they’ll underperform for the first few weeks without context. They don’t know your clients, your standards, your voice, your processes.

Now imagine giving that same person a solid onboarding pack on day one. Your brand guidelines, writing samples, process docs, quality standards. They’d hit the ground running.

Skills are that onboarding pack for Claude. They’re .md instruction files saved in folders that Claude reads automatically. They tell Claude how to work for you.

Instead of re-explaining your preferences every conversation, you write them once. Claude loads the file at the start of every relevant conversation and follows your instructions from the first word.

Here’s the difference in practice:

Skills are the single best thing you can set up in Claude. Everything else in this guide builds on this.

Where Skills live in Claude Code

This guide assumes you’re using Claude Code. It’s where Skills are most useful because Claude Code reads them automatically from your file system. Save a .md file to the right folder and it just works.

Here’s how Skills load in Claude Code:

• Project-level Skills: Save .md files to .claude/skills/ in any project folder. Claude Code reads them automatically when you’re working in that project

• CLAUDE.md files: Drop a CLAUDE.md file in any project root. Claude Code reads it at the start of every conversation. This is where your project-specific instructions live

• Global Skills: Save to ~/.claude/skills/ for Skills that apply across all your projects (like your brand voice)

Skills also work across the wider Claude ecosystem. The same .md file is portable:

Practical takeaway: Start with Claude Code. Build your Skills in .claude/skills/. If you use other Claude products, the same files work there too.

The anatomy of a perfect Skill

Every effective Skill file has three parts. Miss one and the Skill underperforms. Nail all three and Claude actually starts working the way you do.

1. Name and description

The name is how you and Claude identify the Skill. Keep it clear and descriptive. Someone reading just the name should know exactly what it does.

The description tells Claude when to use it. This is the most critical part, and the one most people get wrong. A vague description means Claude won’t activate the Skill when it should.

Here’s the difference between descriptions that work and descriptions that don’t:

The pattern: be specific about the trigger, the scope, and the expected behaviour.

2. Instructions

This is the core of the Skill. Instructions are the rules and standards Claude follows when the Skill is active.

Write them as clear, numbered instructions. Be specific. Be opinionated. The more precise you are, the better Claude performs.

A few principles for writing great instructions:

• Be prescriptive, not descriptive. Don’t say “good formatting.” Say “H2 for main sections, H3 for subsections, bullet points for lists of 3+ items.”

• Include the negative. Tell Claude what NOT to do. “Never use passive voice in headlines” is clearer than “use active voice.”

• Reference other Skills. If your brand voice Skill exists, point to it: “Follow the tone defined in brand-voice.md.” This creates a connected system.

• Add decision rules. “If the document is longer than 2 pages, add a table of contents. If shorter, skip it.” Conditional logic makes Skills smarter.

The most common mistake: Being too generic. “Write well” is useless. “Write in short paragraphs, no jargon, no em-dashes, Australian English, confident but warm tone” is a Skill Claude can actually follow.

3. Examples and context

This is the part most people skip.

Examples show Claude what “good” looks like in your world. Rules tell Claude what to do. Examples show Claude how it should feel. Together, they’re what makes the output sound like you wrote it.

Include 2-3 examples of your actual work so Claude can pattern-match your style. Before/after examples work especially well.

Practical tip: Next time Claude produces output you love, copy it into the relevant Skill file under “Good output.” Next time it misses the mark, add it under “Bad output.” Your Skills get better over time this way.

The 5 Skills every professional should build first

These five Skills cover roughly 80% of knowledge work. Build them in this order. Each one builds on the last.

Skill 1: Brand voice

This is your foundation. Every other Skill references it. Without it, Claude defaults to generic AI tone. With it, Claude sounds like you.

# Skill: Brand Voice

Description: Use this skill for ALL written output. This defines my personal and professional writing voice. Apply these rules to every email, document, post, message, and deliverable unless I explicitly ask for a different tone.

## Voice Characteristics

- Tone: Direct, confident, warm, practical

- Register: Professional but conversational. Write like you’re talking to a smart colleague, not presenting to a board

- Energy: Optimistic and action-oriented. Focus on what’s possible, not what’s wrong

## Writing Rules

1. Use Australian English spelling (organise, colour, analyse, behaviour)

2. Short paragraphs only. 2-3 sentences maximum

3. Never use em-dashes. Use commas, full stops, or “not” instead

4. No corporate jargon: never say “leverage”, “synergy”, “circle back”, “touch base”

5. No filler phrases: remove “I think”, “it seems like”, “in order to”

6. Lead with the point. Don’t bury the insight at the end of a paragraph

7. Use concrete examples over abstract concepts

8. Bold key phrases for scannability in longer documents

9. Use contractions naturally (you’re, don’t, it’s, won’t)

10. End sections with a practical takeaway or clear next step

## Vocabulary Preferences

- Say “use” not “utilise”

- Say “help” not “facilitate”

- Say “start” not “commence”

- Say “about” not “regarding”

- Say “need” not “require”

- Say “show” not “demonstrate”

- Say “build” not “construct”

## Examples

### Good output (match this):

“Here’s the thing about growth. Most founders chase tactics. They want the hack, the shortcut, the one channel that will change everything. But sustainable growth comes from systems. Build the system first, the tactics plug in later.”

### Bad output (avoid this):

“In the rapidly evolving landscape of business growth, it is imperative that founders adopt a holistic approach to their growth strategy. By leveraging synergies across multiple channels and facilitating cross-functional collaboration, organisations can unlock transformative potential.”

Pro tip: Your brand voice Skill should be the first one you build AND the one you update most often. As your writing evolves, update the examples. This is a living document.

Skill 2: Meeting notes

Meetings are where decisions happen but context gets lost. This Skill ensures every meeting summary follows the same structure and makes follow-up effortless.

# Skill: Meeting Notes

Description: Use when summarising any meeting, call, or conversation. This includes processing transcripts, recordings, or handwritten notes.

## Instructions

1. Structure notes: Context > Key Decisions > Action Items > Open Questions > Follow-ups

2. Keep summaries concise. Someone who missed the meeting should get the full picture in 2 minutes of reading

3. Action items MUST include: the task, the owner, and the deadline. If no deadline was stated, flag it as “Deadline: TBD, needs confirmation”

4. Separate facts from opinions

5. Use present tense for decisions, future tense for actions

6. If the meeting had no clear decisions or actions, note that explicitly

7. Include attendee names and roles at the top

Pro tip: Pair this Skill with Claude’s audio transcription. Record your meeting, upload the transcript, and let Claude produce perfectly structured notes in seconds.

Skill 3: Client deliverables

Every document that leaves your desk represents your professional standard. This Skill keeps things consistent across every proposal, report, and strategy doc.

# Skill: Client Deliverables

Description: Use when creating any document intended for a client, partner, or external stakeholder. Includes proposals, reports, strategy documents, and presentations.

## Instructions

1. Open with an Executive Summary (3-5 sentences capturing the document’s value)

2. Clear header hierarchy: H1 for title, H2 for sections, H3 for subsections

3. Table of contents for any document longer than 3 pages

4. Attribution: “Prepared by [Your Name], [Company]” and date on the first page

5. All data claims must include a source

6. Use tables for any comparison of 3+ items

7. End with “Recommended Next Steps” (3-5 actionable items with owners)

8. Format currency as AUD unless otherwise specified

9. Australian English spelling

10. File naming: [Client]-[DocType]-[YYYY-MM-DD]

Skill 4: Email and communication

Email is the most frequent writing task for most professionals. It’s also the most inconsistent. This Skill keeps your tone right while adapting to different recipient types.

# Skill: Email and Communication

Description: Use when drafting, replying to, or editing any email, Slack message, or direct message. Adjust tone based on the recipient type below.

## General Rules

1. Keep emails short. 5 sentences or fewer for routine communication

2. Lead with the point. First sentence tells the reader why they’re reading

3. One email, one ask. Multiple things? Use a numbered list

4. Always end with a clear call to action or next step

5. Never use “Hope this email finds you well” or “Just following up”

6. Sign off with “Cheers,” for colleagues, “Best,” for clients, “Thanks,” when requesting

## Tone by Recipient

- Client: Warm, professional, confident. Restate commitments at the end.

- Colleague: Direct and casual. Skip the pleasantries. Bullet points for updates.

- Cold outreach: 3 sentences max. Lead with relevance. One low-friction CTA.

Skill 5: Research and analysis

Research is where Claude really earns its keep. But without a Skill, you get inconsistent depth and summaries that bury the insight.

# Skill: Research and Analysis

Description: Use when conducting research, analysing data, or synthesising information from multiple sources.

## Source Quality

1. Prioritise primary sources over secondary commentary

2. Prefer recent data (within 12 months) unless historical context is needed

3. Flag any source with commercial bias

4. Never present speculation as fact

## Summary Structure

1. Research Question: Restate what we’re trying to answer

2. Key Findings: 3-5 bullet points with the most important insights

3. Detailed Analysis: Organised by theme, not by source

4. Conflicting Information: Where sources disagree and which is more credible

5. Gaps: What couldn’t be determined?

6. Recommendations: Based on the evidence, what should we do?

Start with these five. They cover 80% of knowledge work. Once you’ve built these, you’ll see exactly how to create Skills for your specific workflows.

What I’ve actually built (and what it looks like in practice)

I want to show you what this looks like when you go past the basics. I’ve built 19 Skills in Claude Code that I use daily. Here’s a sample:

/linkedin-writer, /carousel, /invoice, /podcast-seo, /action-items, /linkedin-post-ideas, /save-context, and a bunch of web scraping and research skills.

Those are just names. Here’s what a few of them actually do.

The podcast transformer

I had a client with 142 podcast episodes on their website. Each page had about 80 words on it. A title and an embed. Basically invisible to Google.

So I built a /podcast-seo skill that pulls the transcript, generates a full SEO content hub with key insights, guest panels, timestamps, and schema markup, then pushes it straight to WordPress. Each page goes from 80 words to roughly 13,000. And it runs the same way for every episode.

The LinkedIn writer

I scraped over 1,000 viral LinkedIn posts and broke down what made them work. The hook patterns, the structures, the formatting, the psychology. All of that got fed into a /linkedin-writer skill with 7 proven post formats baked in.

Now I say “write a LinkedIn post about X” and it picks the best format for the topic, matches my voice, and gives me two drafts in different structures so I can choose. My best-performing posts this year came from that skill.

The personal CRM

I built a system that pulls together my emails, iMessages, WhatsApp messages, LinkedIn DMs, and meeting transcripts into one searchable database.

Before a meeting, I ask “what’s my history with this person?” and get every conversation we’ve had across every channel. What we discussed. Their business updates. What they were working on last time we spoke. No more “sorry, remind me where we left off.”

Ghostbuster

I get a lot of messages across a lot of channels. Emails, iMessages, WhatsApp, LinkedIn DMs. It’s hard to manage all of the messages and things just slip through when you’re moving fast.

So I’m building a system called “Ghostbuster” that scans all of those channels, finds conversations where someone is waiting on a reply from me, and sends me a digest three times a day. It tells me who I owe a response to and on which channel. I’m still refining the workflow, but the fact that I can even attempt something like this with Claude Code and a few Skills should tell you where this is heading.

How to build a Skill in 5 minutes

I’ll give you two paths: the fast way (let Claude build it for you) and the manual way.

The fast way: let Claude Code build your Skills for you

This is how I do it. You don’t need to write Skill files from scratch. Claude Code can read your existing work, spot your patterns, and create the files for you.

1. Open Claude Code in your project folder. If you don’t have a .claude/skills/ folder yet, that’s fine. Claude Code will create it

2. Give it something you’ve already written. Paste in 2-3 emails, a proposal, a LinkedIn post, whatever represents your voice. Then say: “Based on these examples, create a brand-voice.md skill file in .claude/skills/ that captures my writing style, tone, and preferences.”

3. Review what Claude Code produces. It’ll create the file with your voice characteristics, writing rules, vocabulary preferences, and examples. Read it. Does it sound right?

4. Tell Claude Code to sharpen it. This is where you add the stuff it can’t figure out on its own. “Add a rule: never use em-dashes. Add a rule: always use Australian English. Add to the bad output section: anything that sounds like corporate jargon.” Be opinionated. The more specific, the better

5. Test it immediately. In the same Claude Code session, say: “Write a 3-paragraph LinkedIn post about why most professionals are underusing AI tools.” Does it sound like you? If not, tell Claude Code what’s off and it’ll update the Skill file right there

6. Iterate. The Skill lives at .claude/skills/brand-voice.md. Every future Claude Code session in that project reads it automatically

The manual way

If you prefer to write Skill files yourself, open any text editor and create a .md file with the three parts: name and description (when should Claude use this?), instructions (numbered rules), and examples (2-3 samples of your actual work, plus “bad output” to avoid). Save it to .claude/skills/ in your project folder. Done.

Most people find the Claude Code approach faster. It’s often better too, because it spots patterns in your writing that you might not articulate yourself.

Here’s what the same task looks like with and without a Skill active:

Levelling up: debugging and power user tips

When Claude ignores your instructions

It happens. Here’s the quick reference:

Power user tips

Use CLAUDE.md in every project root folder. Claude Code reads it automatically at the start of every session. Drop in your project brief, client requirements, or working preferences and Claude Code picks them up from the first prompt.

Create a “meta-skill” that builds Skills. Tell Claude how you want Skills structured, then say “Create a skill for X” and it follows your format. A Skill that builds Skills. Yes, it’s recursive and slightly ridiculous, but it works.

Version your Skills. When you improve one, save it as v2. Keep the old version so you can roll back. Simple naming: brand-voice-v1.md, brand-voice-v2.md.

Audit your Skills quarterly. Your work changes. Your standards evolve. Set a calendar reminder to review and refine your top 5 Skills every quarter.

Use Skills as training data. When Claude produces great output, extract the patterns and add them to the relevant Skill. When it misses, add the bad output to “avoid this.” They get smarter over time.

Pair Skills with scheduled tasks. In Claude Code, you can set up recurring tasks that run on a schedule. A Skill plus a schedule equals work that happens automatically to your exact standards.

Your first 15 minutes with Skills

Stop reading. Start building.

1. Open Claude Code in any project folder you work in regularly

2. Paste in 2-3 examples of your writing (emails, posts, proposals, whatever represents your voice) and say: “Based on these examples, create a brand-voice.md skill file in .claude/skills/ that captures my writing style.”

3. Review the Skill file Claude Code creates. Does it capture your tone? Your pet peeves? Your formatting preferences?

4. Sharpen it. Tell Claude Code what to add or change. “Never use em-dashes.” “Always Australian English.” “Add this paragraph to the bad output section.” Be specific

5. Test it. In the same session, say: “Write a 3-paragraph LinkedIn post about why most professionals are underusing AI tools.” Check the output. Does it sound like you wrote it?

6. Build your second Skill. Say: “Create a meeting-notes.md skill in .claude/skills/ that structures all my meeting summaries with context, decisions, action items, and follow-ups.”

7. Test that one too. Paste in a rough transcript or meeting notes and see if the output matches your standard

8. Keep going. Add a client-deliverables Skill. An email Skill. Each one takes about 5 minutes. Each one saves you hours

Ready to go deeper? Join the live workshop. 90 minutes, all three Claude tools, hands-on setup. You’ll leave with Cowork configured, your first 3 Skills built, and a Claude Code demo. Two sessions to fit your timezone.

Session 1: Thu April 16, 3pm AEST → Register here

Session 2: Thu April 21, 10am AEST → Register here

This guide was written by Jeremy Cabral, co-founder of Finder (bootstrapped to $100M+ in revenue), now a growth advisor helping founders scale.

If this was useful, share it with someone who could use an extra 8+ hours back in their week.

Subscribe to get the next guide in this series (More Claude playbooks are coming): newsletter.jeremycabral.com

Follow for more: Instagram | LinkedIn | TikTok | X

A highlight reel of the day. the rest of this post is the detailed play-by-play.

5 hours, 43 minutes. It wasn’t a remarkable time, but it was remarkable that I got it done.

20 kilos lighter than when I started training. No injuries. Three days later I was ready to run again.

12 months earlier that outcome made no sense. I was 127.5kg, had moderate to severe sleep apnoea, chronic back pain, sciatica and a VO2 max of 28 ml/kg/min on a treadmill test. I was squarely in the “poor” cardiorespiratory fitness bucket for my age. Around the same time, someone close to me was given what looked like a life-ending diagnosis.

That shock did not turn me into an athlete overnight. But it did light a fuse. I went looking for energy I had never felt before. I did a 10-day water-only fast to prove to them, and to myself, that hard things were possible. I heard about Nedd Brockmann and his run across Australia raising funds for homelessness, and eventually met him at an event for We Are Mobilise. I felt they were a great charity to support I’d been raising funds for homelessness for over 11 years.

Despite managing to run a marathon, I don’t think of myself as a runner. I am a data driven guy who likes to pick goals that look impossible, then use numbers, tools and people to find a way through.

This is the story of how I used 3,049 ChatGPT prompts, a stack of diagnostics and a small F1-team of humans to lose 20kg in four months and go from barely run-walking 2km to running 42.2km.

This is not medical advice. It is a case study in what happens when you point AI at a single near impossible goal and then are willing to do the boring work every day.

Life before: CPAP, caffeine and 3 hours sleep

At Finder’s peak we had close to 600 staff. I was doing 120 hour work weeks in 2021/22, flying regularly, working across time zones, doing calls until 5.30 am, grabbing an hour of sleep, then trying to be present for 13-month-old twins.

On a sleep study I scored 32 apnoeic events an hour and was diagnosed with moderate to severe obstructive sleep apnoea. CPAP became non-negotiable and I managed to get one two days before flying to and working from the US and other countries for 9 months.

My coping pattern was simple: Meal. Coffee. Meal. Coffee again.

If that did not wake me up, energy drink.

If that did not work, more sugar. Then more coffee.

I cycled these just to stay on calls. I woke up feeling like I had been hit by a truck every day. My weight climbed toward 140kg. My back hurt so much that when our girls were born during lockdown in 2020 I had my own bed in the hospital room because the sciatica pain down my leg made it hard to stand, let alone hold them. I was asking for pain killers but unable to get any as I wasn’t the patient, it was brutal. That’s when I first met Yvonne McKenny from Evoker Physiotherapy who helped me get the sciatica under control after several treatment sessions and an exercise program.

I knew the physiology. I read clinical papers for fun at night. I listened to Peter Attia, Andrew Huberman and Will Bulsiewicz on long drives. But my body did not match my knowledge.

It was time to change that.

The 10-day fast and a seemingly impossible goal

The catalyst to getting healthy was not the marathon. It was a combination of factors.

Firstly, I’ve been working with my life coach Craig Hall for years to become the best spiritual, emotional, mental and physical version of myself. This year we were working to help me become more selfish - prioritising myself more so I could turn up better for myself and others in my life. I’ll always be grateful for Craig who helped to lay the foundation to prioritise my physical self.

Secondly, a person close to me had a health scare that looked like a serious diagnosis. This made me realise that if I didn’t get my health in order I could lose the opportunity to be there as my girls grew up was the kick I needed to take action.

I had been reading Valter Longo’s work on fasting and chemo tolerance, and there is emerging evidence that therapeutic fasting can trigger autophagy and help clear damaged cells, although it still needs more study.

Sitting at lunch one day with some people very close to me, I put my fork down and said:

“This is my last mouthful of food for 10 days.”

They stared at me like I had lost it.

For 10 days I drank only water with fasting salts, tracked blood glucose with a CGM and watched my WHOOP strain and recovery like a hawk. By day three I was in deep ketosis. My energy went through the roof. I started apologising to people for how intense I felt in meetings because I was on a keto-high.

That fast did two things:

1. It showed me I could sit with discomfort.

2. It gave me a glimpse of how much energy might be sitting under the extra weight.

A few weeks later I was in Mecca Coffee in Paddington and overheard a guy saying, “I have just been named head coach of the Sydney Marathon.”

That guy was Ben Lucas.

I had met Ben briefly at the opening of his fitness studio Flow Athletic probably a decade earlier. I got his attention.

“Mate, can you coach me to run a marathon?”

He looked me up and down. I was a 125+ kg, sleep-deprived tech founder asking to run 42.2km having never even run a half marathon. He could have laughed me out of the cafe.

Instead he said, “Yeah, we can do it. It will be hard, but you can get it done. Let’s get to work.”

The next week I shuffled 2km around the block with Ben, alternating 60 seconds of running and 60 seconds of walking, and felt like I was dying.

That was the baseline.

I had spent some time really reflecting on what drove me and gave me energy. This marathon was about unlocking that energy to be available to me every single day.

Turning my health into an F1 program

If I was going to pull this off I had to treat my health like we treated growing Finder.

1. Setting a big goal.

2. Having brutal honesty about the starting point.

3. Then a lot of incremental optimisation - or ‘incremental unlocks’ as I commonly say. I might get that tattooed on my arm one day.

I pulled together a little F1 team/pit crew:

• Ben Lucas for marathon coaching and running specific strength work

• Lockeroom Gym for my holistic strength program and weight loss accountability

• Adam Monteith, Yvonne McKenny and Mick Morrison - the team at Evoker Physio to help me avoid injury and optimise my running form.

• In the last month or so prior to race day, a running podiatrist to help me with cadence and orthotics

• My clinical nutritionist Karen Van Esveld at WELLSET plus gastroenterologist and microbiome testing with an Australian lab, Microba.

• The Compounding Pharmacy of Australia and the team constantly supporting my supplementation strategy.

• Dr Will Bulsiewicz (a guest on the Diary of a CEO podcast) to help rebuild my gut after years of IBS.

• NIMBUS & Co for recovery, multiple times a week for saunas and ice baths.

• Craig Hall - my long-term life coach

• A friend Walid Abou-Halloun who offered to run the marathon with me

• Extensive blood panels, plus DEXA scans to track fat, lean mass and visceral fat.

On paper my starting numbers were not great:

• I did a VO2 max test in Chile in December 2024 and finished with a 28 ml/kg/min on lab CPET and mirrored by WHOOP - very low for a 38-year-old male.

• BMI 35.6 with total body fat 32.8 percent and estimated visceral fat area 194 cm².

• LDL 4.0 mmol/L and non-HDL 4.5 mmol/L, both above target ranges, increasing long term cardiovascular risk.

• Vitamin D low at 46 nmol/L, fasting insulin 11 mU/L with HOMA-IR 2.2, nudging toward insulin resistance.

At the same time, the literature is very clear: cardiorespiratory fitness is one of the strongest modifiable predictors of mortality. Going from low to moderate fitness cuts all-cause and cardiovascular death risk by roughly 30 to 50 percent.

So the task became simple:

1. Raise VO2 max.

2. Drop body weight aggressively while protecting muscle mass.

3. Fix my sleep and apnoea.

4. Build a gut that could handle crazy long runs and gels without turning on me.

This is where AI came in.

How I used AI prompting ChatGPT 3,049 times as a control tower

I exported my ChatGPT conversation history and after some data wrangling using Python, managed to chart it too.

I treat ChatGPT like a chief of staff that never sleeps. It knows everything about me and has unlimited context to be able support me 24/7.

Over the 10 months between that first 2km shuffle and the finish line I logged 3,049 prompts in a single project. A few examples of what I fed in:

• Full CPET report with my treadmill stages, thresholds and VO2 data.

• DEXA scans showing fat distribution, lean mass indexes and visceral fat estimates.

• Blood test history back to 2022 including lipids, hormones, vitamin D, inflammatory markers and insulin.

• Gut microbiome results and suggested strains from the lab reports.

• Training plans from Runna and notes from sessions with Ben.

• WHOOP data on sleep, HRV, strain, resting heart rate and step counts.

On top of that I asked it to cross-reference what I was doing with work from people like Attia, Huberman, Rhonda Patrick and Dr B, plus the clinical studies behind them.

Some concrete ways I used AI:

1. Designing a Minimum Viable Training Plan
   I had a finite number of hours to prepare for the marathon. Finder was still a huge time commitment so there wasn’t much spare time to work with. The kids were still young. I prompted ChatGPT to help design a “minimum effective marathon plan” around my CPET thresholds, lab VO2 max and Runna structure. The goal: hit enough volume without getting injured, ruining my joints or nervous system.
   

2. Optimising VO2 max and weight in parallel
   AI built me a simple model: VO2 max is driven by stroke volume, oxygen extraction and body mass. It pushed the idea that dropping body weight is the fastest way to improve relative VO2, especially when you start in the obese range. That guided the decision to target 5kg a month for four months and the guys at Lockeroom Gym were critical in keeping me accountable to this target.
   
   The meta-analysis from Kodama et al showed that each 1-MET (3.5 ml/kg/min) increase in fitness cuts all-cause mortality by roughly 13 percent.
   

3. Periodising nutrition across the week
   I did not have a single “diet”. I had a weekly operating system.

   • Protein-only days on lower training days to drive satiety and preserve lean mass.

   • Carb loading the day before long runs with 300 g plus of carbs, then Maurten gels every 25 minutes while running.

   • Occasional fasts (24 - 36 hours) to re-set appetite and get deeper into fat loss.
     

4. AI worked out macro ranges and timing templates for strength days, VO2 sessions, tempo runs and long runs, then I had Karen and Dr B sanity-check them.
   
   Clinically, losing 5 to 10 percent of body weight improves blood pressure, insulin sensitivity and lipids. For sleep apnoea, 10 to 15 percent weight loss can halve severity.
   

5. Gut rebuild plan
   Since about October 2024 Dr B hammered one simple rule into me: 30 plus different plant foods a week to diversify the microbiome. That target comes from the American Gut Project, where people eating 30 or more plants per week had significantly higher microbial diversity than those eating 10 or fewer.
   
   I worked with Karen from WELLSET to develop a targeted supplementation strategy to support my gut repair.
   

6. Sleep, CPAP and thermoregulation
   
   I gave ChatGPT my sleep study, CPAP settings, Eight Sleep temperature data and WHOOP recovery scores. We iterated on a “sleep protocol” that tried to hit 7.5 hours a night, with strict caffeine cut-off times, magnesium glycinate and threonate, a GABA/taurine/inositol compound from my Sydney pharmacy, and a consistent wind-down.
   
   Restoring sleep was not optional. Poor sleep worsens insulin resistance, appetite and accident risk, and CPAP plus weight loss can meaningfully improve apnoea.
   

7. Treadmill vs road and injury risk
   
   I did almost all my early running in preparation for the marathon on a treadmill. Adam Monteith gave me this brilliant advice and it was key to me not getting majorly injured in the lead up. AI pulled studies showing that treadmill running at 1 percent incline gives similar metabolic cost to outdoor running but with roughly 40 percent lower impact forces at the knee and ankle, which matters when you are 120+kg.
   
   I only moved to regular road long runs about eight weeks out from race day.
   Adam and Yvonne were on speed dial the moment I had a niggle and managed to step in within hours to deal with it and give me strategies to evade any risk during my runs.
   

8. Race strategy in real time
   
   During the race I wore SHOKZ bone-conduction headphones with a metronome app ticking in the background at my target cadence. Several times I pulled out my phone, dictated into ChatGPT and asked things like:

   • “I am 26km in. Pace 7:30/km. HR mid 150s. Legs ok. Stomach ok. What should I adjust for the next 5km?”

   • “I am starting to see people walking. I feel strong. What is the risk of pushing now vs waiting until 35km?”

It did not give magic secrets. It reminded me of basic race craft: do not surge because you feel good at 25km, keep fuelling, respect your heart rate caps, save the final push for the last 30 minutes. When your brain is fried, having a calm, data-driven voice in your ear helps.

The sleep stack

Optimising my sleep was critical to get this done. I had already been focused on this for a couple of years but went to another level several months prior to starting the marathon. I even live-streamed my sleep from the Finder office to bring attention to how I’d been optimising my sleep.

Here is the tame version of my supplement stack. DM me on LinkedIn if you want a full list of everything else I take.

My sleep stack:

• Magnesium Glycinate

• Magnesium Threonate

• Zinc Picolinate

• CALM

• Ashwagandha

The four-month fat loss block

The headline number people latch onto is “20kg in four months”.

The raw stats:

• Start: 127.5kg in April.

• Lowest: 107.5kg in August, just before the marathon.

That is about 1.25kg a week. Aggressive, but not crash-diet territory for someone starting at my weight.

Here is what those months actually looked like.

Month 1 - Shock and data

Goal: Stop the bleeding.

• No alcohol. I made a few exceptions along my journey but otherwise zero.

• Time-restricted fasting when I could and three protein-heavy meals a day.

• 3 to 4 training sessions a week plus 3 short run-walks.

• CPET, DEXA, full bloods, microbiome test.

Weight came off fast in this block, a mix of glycogen, water and early fat loss.

Month 2 - Structure

Goal: Set a rhythm I could hold.

• Follow Runna plan with Ben’s tweaks: 1 interval/VO2 session, 1 tempo, 1 long run, 1 to 2 easy runs. Some weeks I could only squeeze in two runs, and it was enough to get me through.

• Strength on non-running days, focused on posterior chain, glutes, hamstrings and calves.

• Protein first at every meal. Aiming for 30+ plants a week.

• Two higher-carb days a week preparing for the long run.

• 10,000 steps minimum every day.

My WHOOP strain scores on long run days were on some days the highest of anyone in the Sydney group on WHOOP. A colleague screenshotted the leaderboard and called me insane.

Month 3 - Hard things

Goal: Become the kind of person who does hard things on purpose.

• Treadmill for all runs

• Hill repeats, tempo blocks and VO2 Max optimising intervals

• 24 to 36 hour fasts to move into deeper fat loss.

• Cold plunges: 15 minutes at Nimbus immediately after long runs.

There is a large Finnish cohort showing that 2 to 3 saunas a week cut fatal cardiovascular disease risk by around 25 percent, and 4 to 7 a week by about 40 to 50 percent, so sauna and heat stress were in the mix as well. Also that sauna usage can have a positive impact on VO2 Max.

Month 4 - Race specificity

Goal: Arrive at the start line light enough, fit enough and uninjured.

• Outdoor long runs up to 29km around Narrabeen Lagoon, and then my final 32km long run in torrential rain that finished in the dark at Centennial Park.

• I had to do 5 half marathon efforts in my preparation for race day fueled with Maurten gels every 25 minutes plus electrolyte drink (thank you Hyro). Studies on nitrate rich beetroot juice are mixed but suggest small benefits for short time trials, so I added beet shots and tart cherries around key sessions to help blood flow.

• Cadence work with a metronome app beeping in my ear. I had a running gait analysis in the city where the podiatrist told me to lose the metronome so I would not go mad. I ignored that advice in most of my running and had it on as often as possible to help prevent injury. I remember turning them off after runs and still hearing the beeping for minutes after.

I practised running with music I disliked but with the right beat, then saved my favourite punk and metal tracks for the last 30 minutes of big sessions. It was like a Pavlovian reward.

Race day: metronomes, Nedd Brockmann and AI

Race day started crossing the Sydney Harbour Bridge. The weather was perfect.

For the first 30km I stuck to the script:

• Maurten gel every 25 minutes.

• Water at every aid station.

• Metronome ticking away beneath a podcast or bad pop playlist.

Around 32km you start seeing people hit the wall. They cramp, they shuffle, they walk. I saw it happening in front of me. When you are still running at that point it is a mental trap. Your brain says, “Everyone is walking. It is ok for you to walk.”

Somewhere around there I heard, “Go Jeremy!” with a few expletives included and looked up to see Nedd Brockmann on the sidelines. I had only just met him at a charity event for We Are Mobilise a few days before. Him yelling my name, and after a quick selfie together it flipped a switch and I started pushing again.

The last 10km were delirious. I was singing out loud. My physio Yvonne saw me and asked what on earth I was doing. I yelled back, “I am pumped, baby!”

In the last 2km I started to see familiar faces. Friends and family. Then my daughters. They were there around Mrs Macquarie’s Chair, running along next to me for a bit with signs in their hands encouraging the runners.

I was coming into the home stretch in immense pain and totally missed a group of my friends who were yelling my name out. I just kept telling myself to not stop and I wanted to be crossing that finish line running.

I crossed the line, walked the stiff-legged finish chute wanting nothing but a can of Coke, hugged my wife and burst into tears. After seeing friends and family briefly I went straight to NIMBUS & Co and did an ice bath for 15 minutes straight. I did that for the next 3 days after the race too and it dramatically sped up my recovery.

WHOOP had my strain score maxed. But three days later I was back running easy laps, no injuries. I put that down to the incredible running coaching I was given and the support from the EVOKER team.

The stats: the health and fitness transformation

Here’s how I managed to move the needle before and after completing the marathon:

Life after: bio age, girls and boring consistency

Since the marathon, I kept up the running at one to two sessions a week. I now have set my sights on optimising my blood markers, my bio age and getting as healthy as I can by the time I turn 40 in January.

My weekly program looks like:

• Three strength sessions a week.

• One VO2 or interval session.

• One longer run or race practice.

• 12 to 15,000 steps most days.

• Sauna three to four times a week.

My sleep optimisation is still locked in.

My biological age on WHOOP is dropping toward my actual age. I am pushing toward sub-100kg and ideally the mid-90s at 187 cm, which is a much healthier range.

The best part has nothing to do with numbers.

My girls now ask, “Dad, are you going for a run so you can be healthy?”

They came to meet me at the end of one long run near the NSW Art Gallery and dropped their scooters to run the last stretch with me. I bought them little joggers and they think they have “race shoes” like Dad.

When your five-year-old says, “Dad, your tummy is not big anymore,” it hits harder than any lab result.

In saying that, I am super interested in seeing how they come up with my upcoming blood test with Everlab.

What you can steal from this experiment

You do not need 3,049 AI prompts, a VO2 lab test and an Eight Sleep to make progress. Here are the pieces I would keep if I had to strip this back for a normal human with a job and kids.

1. Pick an “impossible”, emotional goal
   “Lose weight” is vague. “Run the Sydney Marathon next August” is clear. Tell everyone. Let their doubt annoy you into action. (Thank you Doone and P-A for pushing me to have an ambitious goal like this).
   

2. Get some real baseline data
   At minimum: weight, waist, blood pressure, fasting lipids and glucose, and some sense of fitness (e.g. a 1km time trial). If you can afford DEXA, a VO2 test and a sleep study, even better. It is easier to change what you measure. In all scenarios a health tracker like WHOOP will give you plenty of data to work with.
   

3. Use AI as an assistant, not a guru
   Let ChatGPT summarise papers, draft training week templates, explain VO2 or sleep apnoea in plain language and suggest questions for your doctor or coach. Always run advice past real clinicians and coaches who know your history (and can tell you when ChatGPT is hallucinating.)
   

4. Build your own F1 team, one person at a time
   I did not hire everyone at once. I added people as I hit bottlenecks: a physio when I kept almost getting injured, gut support when my IBS flared with gels, specific exercise prescription when my knees hurt.
   

5. Sleep like it is your main job
   If you snore, stop breathing at night or wake exhausted, push hard for a proper sleep assessment. Fixing my apnoea changed everything.
   

6. Raise fitness and cut weight together, but respect the maths
   Aim for 0.5 to 1.0kg of loss a week. Use protein and fibre to stay full. Use carbs around hard sessions and tactical fasting to manage calories over the week.
   

7. Feed your gut 30 plants a week
   Diversity, not perfection. Frozen veg, herbs, spices, nuts and seeds count. Your microbiome will thank you and so will your long runs if you have a sensitive gut like me. Gels caused me a lot of issues in my preparation.
   

8. Do strength work even if you only care about endurance
   Sarcopenia and low muscle mass are a straight line to frailty. And stronger legs handle more running load. Strength work is also a key to longevity. Some cohorts show weak quadriceps or poor chair-stand performance raising death risk by about 50- 65%.
   

9. Accept that most of the work is boring
   Sleep. Steps. Cooking. Logging runs. You do not see those in headline stories but that is where the change happens.
   

10. Make it about people, not just performance
    I did not chase a marathon to post a medal on Instagram. I did it because I want another 10 good years with my girls, and because I want my body to match the health knowledge I have spent years absorbing. I was chasing a new level of energy that could be available to me every day, and I found it.

If you are reading this as someone who feels stuck, overweight, exhausted and a bit embarrassed that your body does not match what you know, I get it.

You do not fix that in one weekend.

You do not need to.

You just need one stupid goal, one clear why, and the willingness to let AI, good clinicians and a boring training log support you while you do the actual work.

Now it’s your turn to make the impossible possible.

Looking back, I’m proud to have played a part in three milestones that stand out in my mind…

1. Being a part of a company that bootstrapped its way to $100m+ revenue.

2. Expanding into 20+ countries helping over 35+ million people a year.

3. Supporting 1,100+ Finder crew to achieve their full potential at Finder (and beyond).

I have always obsessed over building a better product - chipping away every day at another guide to educate, building another comparison, leaving no stone unturned in our quest to help people better understand the products they were considering and improve their financial decisions.

Over time, those incremental ‘unlocks’ every day led Finder to being a brand trusted by millions globally. This of course was only possible because of the remarkable crew that joined us on this mission.

I had the privilege of leading and working alongside our crew for so long, who passionately worked on something they knew would be bigger than themselves. We built incredible camaraderie, friendship and have overcome every challenge that’s come our way, building lasting impact that none of us could have created alone.

We set ambitious goals, and always found ways to achieve them no matter how impossible they seemed. Our successes are marked by the medals honouring our achievements on our office walls, and just as much by how we have risen from setbacks, stronger each time.

I have loved deeply the complexity of growing and scaling Finder. I’ll still continue to provide my guidance on the road ahead as a Special Advisor, shareholder and in board meetings.

Thank you Fred and Frank for the wild ride and our “FFJ” trilogy and partnership. We have built such an amazing company together.

I’m proud to leave Finder in your hands, and as you lead the company forward - I’ll always be cheering for you and the crew through the many ‘Go Live! 🚀’ moments ahead.

To all of our customers, partners and crew - it’s been an honour serving you and building Finder together. You allowed me to contribute something meaningful to this world that genuinely helps people and has helped me deliver on my life’s purpose.

For now, I’m looking forward to spending more time with my family, voraciously learning, and and as I have time to reflect, sharing what I pick up along the way.

We are all experiencing such a transformational moment and I’m grateful to have the time to continue immersing myself in AI and how it will continue to push the limits of what is possible.

I have endless stories, highlights, and lessons from my time at Finder, and I want to give back by continuing to share them to honour the Finder crew and everything we have done together.

For anyone that’s been on the journey with me, I’d love to hear your favourite Finder moments in the comments.

With endless gratitude and love,
J 💙