Not a developer. 17 things shipped since March.

On 10 March I created my first repository. Today is 29 July. There are 17.

I’m not a developer. I’ve never been a developer. I didn’t learn to code this year, and if you put me in front of a blank file and asked me to write a React component I’d still be sitting there tomorrow.

What changed is that I stopped needing to.

The list

Not a portfolio. Just what’s actually there, in the order I built it.

A promotions dashboard for Get Lucky that reads live data out of Google Sheets. A certificate product called Fairways to Heaven. Golf Day Pro. Subscription landing pages with a PayFast webhook wired into them — the first time I’ve owned the payment plumbing in one of my own businesses instead of paying someone else to own it for me. The Get Lucky marketing site. A site for Epic Golf, the indoor Trackman simulator in Cape Town. A microsite for the Indwe partnership. Ever After Returns. Sold Direct. REKRD, then a coaching layer on top of REKRD. And this website.

There’s also one called claudefit, a small game that tells you to get up and stretch while the AI is working. I built it in an evening. It is not a business. I mention it because the ability to build something silly in an evening is the same ability that lets you build something serious in a weekend, and I don’t think people take the second one seriously until they’ve done the first.

Four and a half months. Seventeen repositories. One person who can’t code.

Excel stopped being a specialist skill

This is the one I’d lead with if you only read one section.

Excel was the gate. Properly modelling a business — cost of sales, channel margins, distribution, headcount, a sensitivity analysis on the three drivers that actually matter — was something you either had a finance background to do, or you paid someone. I’ve sat on both sides of that. I’ve paid for models, and I’ve built them badly myself at 1am.

I charge R15,000 to build a founder a 5-year cashflow projection. It’s a fair price. It takes days and it holds up in a room full of investors.

With Claude working directly in Excel, I now build feasibility studies and projections in minutes. Not a rough sketch — a real model, with the formulas intact, that I can interrogate and defend.

That should probably worry me more than it does. What it actually means is that the modelling stopped being the expensive part, and the thinking became the expensive part. Which is where it should have been all along. A model is only as good as the assumptions you put into it, and no amount of AI will tell you whether your retailer is really going to take 38% or whether you’ll actually ship in March.

But the mechanical work — the part that used to keep most business owners out entirely — is gone. If you’ve avoided building a proper financial model because you don’t trust yourself in a spreadsheet, that reason has expired.

Thirty-two storefronts, one template

Get Lucky runs the Hole-in-One Challenge across a growing list of golf clubs. Every club wants its own branded page — their colours, their logo, their course photo, their name on it. That’s reasonable. It’s also historically the thing that makes adding a club expensive enough that you think twice.

There are now 32 of them. Clovelly, Zimbali, Sun City, Mount Edgecombe, Wild Coast, Simola, San Lameer, Glendower, and the rest.

Each one is a config file. Club name, logo, primary and secondary colour, hero image. That’s it. The subscription flow and the payment webhook underneath are the same code for all of them.

Adding a club used to be a project. Now it’s a file, and I can do it between meetings.

That changes something more important than my calendar. When adding a product costs almost nothing, you stop rationing. You stop having the meeting about whether this one is worth the build. You just add it and let the market tell you.

The agent that answers first

Sold Direct is a property business — sell your home privately, keep the commission. It runs on WhatsApp, because that’s where South Africans actually are.

Behind it is an agent that handles the conversation. It knows the process from mandate to registration, answers what it can, and drafts what it can’t for a human to approve before anything goes out. It’s wired into the real WhatsApp API, it has its own knowledge base and tools, and it has evaluation tests so I can tell whether a change made it better or worse rather than guessing.

I want to be careful here, because “AI agent” is doing a lot of lying in the market right now. This is not a robot running the business. A property transaction has PPRA-registered practitioners in the loop and legal obligations that do not bend. What the agent does is take first contact — the same twenty questions, asked at 9pm, that used to either wait until morning or cost me a person.

That’s the honest version of agents for a small business. Not replacing the team. Handling the first reply so the team gets the second one.

Teaching it my rules

The most useful thing I’ve done isn’t on that list of 17 at all.

Get Lucky runs campaigns with Indwe Risk Services, and every mailer carries compliance wording their compliance team signed off. Quote-only mechanics. Specific phrasing that cannot drift, because if it drifts we’re making a claim we aren’t licensed to make.

That used to live in my head and in a document nobody opened. Every new campaign was a fresh chance to get it wrong.

So I wrote it down as a rule the machine follows. Now when I say “make me a mailer for the Wild Coast Classic,” it starts from the approved structure, keeps the compliance wording exactly as signed off, and asks me only what actually changes — venue, dates, the offer line, the photos.

That’s not automation. It’s institutional memory. The rule can’t be forgotten by whoever is doing the job at 11pm, because they’re following a rule that can’t be skipped.

Every business I’ve built has had five or six of these. Rules that live in the founder’s head and leak the moment anyone else touches the work. I’ve started writing them down.

Where it falls over

I’d be selling you something if I stopped there.

It is confidently wrong sometimes. Not often, but often enough that you can’t switch off. If you don’t know enough to smell when something’s off, you’ll ship the mistake. That’s not a reason to avoid it — it’s a reason to stay in the room.

It won’t tell you what to build. It has no opinion about whether your business should exist. Every one of those 17 repos started with me deciding the thing was worth making, and that part hasn’t been automated and won’t be.

The first two weeks were bad. I was asking questions and getting thin answers, and I nearly wrote the whole thing off. What fixed it wasn’t a trick. It was realising I was asking when I should have been briefing — the same skill I’ve used for 15 years on designers, distributors and investors, pointed somewhere new. Nobody told me that. I wish they had.

What it actually means

I keep hearing that the first one-person billion dollar company is coming. Maybe. I don’t know.

What I do know is more boring and more useful: the gap between having an idea and having the thing has collapsed. Not to zero. But from months and a budget, to a weekend and no budget.

For an operator that’s the whole game. Most of my last decade was spent choosing which ideas to kill because I only had the resources to chase one. That constraint is materially different now, and I don’t think most business owners have noticed yet.

Seventeen things since March. None of them by a developer.

I run a 3-hour workshop teaching exactly this to business owners and executives. Eight seats, in person at Ideas Cartel in Claremont. You bring a laptop and one real job from your business, and you leave with it running — not a folder of notes about AI.