I Went From Talking About AI to Building With It in 5 Weeks. Here's What Changed.

The Thing Nobody Tells You About Learning AI

I ran Startmate - Australia's biggest accelerator. I invested in 140+ companies. I know how to evaluate technology.

So when AI exploded, I did what every smart operator does.

I researched.

I attended the AI Roadshow in mid-2025. Five cities. Great speakers. Polished decks. Everyone talking about potential.

It was useful. I learned the landscape. I understood the models.

But it was all presentation. No building.

Here's what I didn't realise: watching someone present about AI is completely different from watching someone build with it.

Presentation is theory. Building is reality.

Theory makes you think. Reality makes you move.

When my friend opened Claude Code and started typing, I saw something I'd never seen in any conference:

• He made mistakes. Claude caught them.
• He asked dumb questions. Claude answered without judgment.
• He changed his mind halfway through. Claude pivoted with him.

It wasn't perfect. It was productive.

And that's when I realised: I'd been learning about AI instead of learning with AI.

Day One Was Terrible (And That's Why It Worked)

I left that office session buzzing.

Got home. Opened Claude Code. Sat down to build.

And immediately fell on my face.

I experimented too much. Asked terrible questions. Tried to build three things at once.

My first agent took four hours and barely worked.

But here's what happened:

Even when I was asking bad questions, Claude was teaching me how to ask better ones.

I'd type: "Build me a script that does X."

Claude would respond: "Here are three approaches. Option 1 is simplest but limited. Option 2 scales better. Option 3 gives you the most control. Which fits your use case?"

I wasn't just getting code. I was getting context.

By the end of day one I'd learned:

• How to structure a question
• How to give Claude constraints
• How to iterate without starting over
• How to ask "why" without feeling stupid

The unlock wasn't the tool. The unlock was permission to ask more questions.

In six months of conferences, I'd asked maybe ten questions.

In one day with Claude Code, I'd asked hundreds.

And every question made the next one better.

The Shift From Consumer to Builder

I didn't ease into this.

The day I started building, I became obsessed.

I ran out of tokens every single day for two weeks.

I'd wake up at 5am with an idea for an agent and be three hours deep before breakfast.

My wife Gabby would come into the office and find me staring at logs, debugging a Todoist integration, muttering "just one more test."

This wasn't discipline. This was dopamine.

I've always loved building product. It's the part of startups that lights me up.

But for 15 years, building meant:

• Finding a technical co-founder
• Writing specs
• Waiting for dev cycles
• Managing scope creep

With Claude Code, all of that disappeared.

I could go from idea to working prototype in an afternoon.

No bottleneck. No translation layer. Just building.

Here's what I shipped in five weeks:

None of these are perfect. All of them are useful.

And every single one taught me something I used in the next build.

That's the shift.

I stopped being someone who talked about AI.

I became someone who shipped with AI.

What Actually Blocked Me (And What Unblocked Me)

For six months I told myself I didn't have time.

"I'll learn this when things slow down."

"I need to finish X first."

"It's too complicated to start right now."

All lies.

The truth? I didn't start because I didn't know where to start.

And the cure wasn't a course. Or a tutorial. Or a roadmap.

The cure was a friend sitting me down and saying: "Build this. Right now. I'll help."

He didn't give me theory. He gave me a project.

"Let's build a script that pulls your calendar events and formats them into a daily email."

Small. Concrete. Useful.

We built it in 90 minutes.

And that one session did more than six months of reading ever could.

Because it gave me proof.

Proof that I could build.

Proof that the tool worked.

Proof that the barrier was in my head, not in the technology.

Here's what I wish someone had told me:

The first step isn't the hardest. The next step is the hardest.

Starting is easy when someone's sitting next to you.

The hard part is opening Claude Code the next morning and building alone.

That's the moment most people stop.

I almost did.

What kept me going? I gave myself permission to build badly.

My second agent was a mess. Hardcoded paths. No error handling. Broke twice before it worked.

I shipped it anyway.

Because shipping a bad agent teaches you more than planning a perfect one.

And once you've shipped two things, the third gets easier.

By week three I wasn't thinking "Can I build this?"

I was thinking "How fast can I build this?"

The Opportunities I Missed (And the Ones You're Missing)

I started building in January 2026.

Claude Code launched mid-2025.

That's six months I didn't use it.

Six months of:

• Manually writing LinkedIn posts
• Copy-pasting meeting notes into emails
• Managing my Todoist by hand
• Running reports I could have automated

I think about this a lot.

How many hours did I waste?

How many ideas didn't I ship because I thought "I'll need a developer for that"?

How many small problems did I live with because solving them felt too hard?

Hundreds.

I sold Startup Striders in an all-cash deal that started with a LinkedIn post.

Imagine if I'd had the LinkedIn Content Engine six months earlier.

Imagine if I'd been shipping posts consistently, in my voice, at scale.

How many more opportunities would I have created?

I don't say this to beat myself up.

I say it because you're in the same position I was six months ago.

You're reading this instead of building.

You're thinking "I'll start when I have time."

Or "I'll start when I understand it better."

Here's the truth:

You'll never feel ready. Start anyway.

Every day you wait is a day you're solving problems manually that you could automate.

Every week you delay is a week you're not learning.

Every month you postpone is a month someone else is shipping faster than you.

I'm not saying this to pressure you.

I'm saying it because I wish someone had said it to me in July 2025.

What I'd Tell Myself Six Months Ago

If I could go back to mid-2025, here's what I'd say:

1. Stop learning about AI. Start learning with AI.

Close the tabs. Stop reading threads. Open Claude Code and ask it a question about your actual work.

The fastest way to learn is to use it for something real.

2. Build the smallest useful thing.

Don't start with "I want to build an AI agent that runs my business."

Start with "I want a script that emails me my calendar every morning."

Small wins compound. Big ambitions paralyse.

3. Ask more questions than you think you should.

Claude doesn't judge. It doesn't get impatient. It doesn't think you're dumb.

Ask why. Ask how. Ask "what if."

The quality of your output is directly tied to the quality of your questions.

4. Ship bad code.

Your first agent will be terrible. Your second will be better. Your tenth will be good.

But you'll never get to ten if you don't ship one.

5. Find someone who's already building.

Theory is useful. Watching someone build is transformative.

Find a founder, a friend, a community member who's shipping with AI.

Watch them work. Ask them questions. Build alongside them.

That one session will save you months.

6. The blocker is never time. It's clarity.

You don't have an AI problem. You have a starting problem.

Pick one task you do manually every week. Automate it. Today.

That's your entry point.

What Happens When You Actually Start

Five weeks in, I'm running 15+ agents.

My Chief of Staff handles my tasks, calendar, and email.

My Monday Standup briefs me with seven AI department heads every week.

My LinkedIn Content Engine writes posts in my voice using 4.1M words of my past content.

I ran the first Build Hour live on LinkedIn. 90 people watched me build an agent in real time.

I'm spending $250/month on Claude Code, $100/month on Vercel, and it's the best ROI in my business.

None of this happened because I got smarter.

It happened because I started.

And here's what changed:

I'm not waiting anymore.

I used to think "I'll build that when I have time."

Now I think "I'll build that this afternoon."

I used to think "I need a developer for that."

Now I think "I need 90 minutes and Claude."

I used to think "That's too complex."

Now I think "That's three agents and a database."

The shift isn't technical. It's psychological.

I went from someone who consumed AI content to someone who creates AI systems.

And the gap between those two states is smaller than you think.

It's one session. One project. One decision to stop learning about it and start building with it.

That's it.