One session. That is how long it took to go from an empty file to a live blog on 11 June 2026.
People assume a site like this needs a team, and mine had one. A Google engineer who cared about search and speed. A typography designer from Apple and MIT. An engineer from Tesla and Caltech who was fussy about motion. And a senior Microsoft product manager whose job was to keep the other three honest on accessibility and quality.
None of them exist. I never emailed them, paid them or sat through a meeting with them. I wrote them into a brief, gave the brief to Claude, and Claude built the site. This post is about why that worked better than I expected, and about the parts that did not.
Four job titles instead of four salaries
I did not ask for a nice blog. A request like that gets you a nice blog, which is another way of saying a forgettable one. Instead I described the people who would be embarrassed if the site was bad.
It works because a named standard is harder to wriggle out of. If one person in the room is obsessed with page speed, then an external framework is a failure, not a style choice. If another person owns typography, then a random font is a failure too. The roles turned vague taste into things that could be checked.
| The imaginary person | Their job | What they left on the site |
|---|---|---|
| Google engineer | Search and performance | Structured data on every post, no external CSS frameworks |
| Apple and MIT designer | Typography and design system | Newsreader for headings, Inter for body, IBM Plex Mono for labels, an 8 point spacing grid |
| Tesla and Caltech engineer | Precision and animation | Every animation uses only transform and opacity, so nothing makes the page jump |
| Microsoft product manager | Accessibility and quality | A skip link, labelled images and charts, and a reduced motion setting that switches animations off |
Read the last column again. Each imaginary person turned into a rule I can check. I did not plan that part. I thought the roles were a trick for getting a good first draft. It turns out the rules they left behind are what keep the blog consistent five months later.
The boring choiceOne file, no dependencies
The whole blog is hand-written HTML and CSS. No Bootstrap, no Tailwind, no jQuery, no build step. Every post is one HTML file with its styles inline. I know that sounds old fashioned. I also think it is the best decision in the project.
There is nothing to update, so nothing breaks while I am not looking. The pages are small, so they load fast. The cost is real, though. Every post carries its own copy of the styles, and I have to keep them in step by hand. I accept that trade every time.
Write the standards down before you write the first post. Mine live in one reference document: colours, fonts, spacing, performance rules and writing rules. Every later change starts by checking that page.
What Google's PageSpeed test says
A fast site is easy to claim and hard to prove, so I ran one of my longer posts, the d-Matrix piece, through Google PageSpeed Insights on 15 November 2026. It is a lab test, so the result does not depend on how many people visit. On desktop the page scored 90 for performance. The first content appeared in 0.7 seconds and the largest element in 1.8 seconds. Layout shift was 0.002 and total blocking time was 110 milliseconds.
The scores I care about most are the ones the brief asked for. Search and best practices both came back at a perfect 100, and accessibility at 96. That is the Google engineer and the Microsoft PM earning their imaginary salaries.
Mobile is the weaker side, and I would rather print the number than hide it. On an emulated mid-range phone with slow 4G, the same page scored 74 and the largest element took 6.8 seconds to appear. The report flags render-blocking requests and oversized images as the first things to fix, so that is my next job.
This is one page, tested once, in a lab with a deliberately slow connection. Scores shift by a few points between runs, so treat the pattern as the useful part: strong on structure and accessibility, good on desktop, with room to improve on mobile.
Posts first, polish later
The first post went up the next day, 12 June. It was a guide to watching the FIFA World Cup 2026 from India. By the end of that month there were about a dozen more. Where Guadalajara got its name. How far a swimming lap really is. Ethanol and sugar stocks. What a BMW X1 rear shock absorber costs in India. Whether AI is a bubble like 2008. Why Argentina went from rich to broke.
That is a strange list, and I like it. The tagline is Beyond One World, Beyond One Idea, and the archive shows I meant it. Later the site grew a workshop side as well, with a TradingView indicator called AlphaEdge and a debt payoff calculator, built in the same plain HTML.
What I actually thinkA persona is a checklist with a name
I want to be straight about this. Nobody from Google, Apple, Tesla or Microsoft has looked at my site. The imaginary team was a way of asking for high standards. It was never an endorsement. If you read "world class" in the title as a claim that experts approved the blog, it is not that. It is a claim that I set the bar high and then checked the work against it.
Who would be embarrassed if this page was bad, and what would they check first?
The answers become rules. Rules outlive the conversation.I think we undersell this method. Most people prompt for a result. I got better work by prompting for a standard, and the standard is what I still use every week.
And yes, an AI built this blog. I am saying so here. It would be odd to write a post about honest standards and then hide who did the typing.
If you are building your own site, write the team first. Then write down what each person would refuse to ship. That list is your real design system.
Common questions