It starts with me talking. Voice memos in the car, on a walk, at the desk — whenever the thought is alive. I don't write first. Writing too early flattens things; speaking first keeps the texture. A lot of what ends up on this site came out of a ten-minute voice note that surprised me while I was saying it.
I use Antigravity for this — it transcribes, organises, and lets me build on a voice library without losing the thread. Before Antigravity, it was scattered across three apps and usually lost.
The transcript becomes a prompt. Sometimes the prompt is a question — what is this really about? Sometimes it's a structure request. Sometimes I'm testing whether an idea holds when I have to explain it to something that doesn't nod along to be polite.
I use Gemini and ChatGPT here for thinking and structuring — long-context reasoning, first drafts, pushing back on my framing. These tools are good at helping me find what I was trying to say.
Once the content is shaped, I hand the build to agents. Pages, code, HTML, CSS, file structure — all of it. I describe what I want; the agents build it. I run most of the site build through Claude in Cowork mode — which is what made it possible to go from ideas to live pages without ever opening a code editor.
For coding tasks where I want inline edits, I use GitHub Copilot. For reasoning through architecture or edge cases, I'll reach back to Gemini. The right tool for the right pass.
This part never gets delegated. I read everything before it goes live — not just for typos, but for whether it sounds like me. AI-generated prose has a flatness to it if you're not careful. I'm looking for that and correcting it. I'm also checking that nothing has drifted from what I actually believe.
Sometimes I'll rewrite whole paragraphs by hand. Sometimes one word is wrong and it's the most important word on the page. The review pass is where I earn the right to put my name on it.
A git commit and a push. The site deploys via a webhook to Hostinger. No build step, no CI pipeline — just static HTML that loads fast on slow connections. There's something satisfying about that simplicity after a decade of over-engineered stacks.
The whole pipeline from voice note to live page can happen in a single afternoon if the idea is clear. Most things take longer because the idea isn't clear — and that's fine too. Clarity earns its time.
Want to see this on a single real thought? Unedited takes one voice note — the raw audio, the AI-cleaned version, and the final piece — and lays all three side by side.
I'm a computer engineer. Three distinct phases — and I've lived all of them consciously.
First, developer. Parts of it I loved; parts of it I didn't. I was always more interested in the problem than the syntax. Then consultant — helping businesses adopt platforms, build platforms, change the way they thought about technology. Then Global Technology Director — building large-scale tech teams that actually deliver: agile, lean, test-driven, continuous deployment. I'm a big believer in the last one. Ship small. Ship often. Learn fast.
And now — high performance leadership and teams. Helping people and organisations become more than the sum of their parts. That's where most of my energy has been.
Then the AI revolution arrived, and I found myself at a crossroads. Reinventing, again. The question I keep sitting with: what does it mean to be a builder in a world where AI builds?
This workflow is my current answer. It's imperfect and evolving. But it let a computer engineer — who had spent two decades managing engineers rather than writing code — come back to making things with his hands. That felt worth writing down.
The messy bits. The half-built pages that never went live. The ideas that seemed clear in a voice note and collapsed as soon as I tried to structure them. The times I've re-written a page three times because the first two weren't honest enough.
The pipeline above describes the happy path. Most of the actual work is the part before that — figuring out what I want to say, and whether it's worth saying.