A red open-wheel single-seater race car taking a corner on a race track in low evening light, its driver in a dark closed helmet.

Case studies

AI is the race car. Your website still needs a driver.

By Luis Gallego

October 9, 2026 6 Min Read

Image generated with AI. Art direction: Grafica.

The two sites compared in this post are both live. The finished site: netica.solutions. The starting line, built in one day and kept online on purpose: 152f2349.netica.pages.dev.

If you are about to build or rebuild your website, three things are true this year.

It will be fast. AI tools turn a brief into a working website in days. Most teams quoting you have them.

It can look like the others. Grids of identical cards. Strips of big numbers. Blocks that float up as you scroll. You have seen that page many times.

It can pass every test. Google’s test can give it 100. That score says the page is correct. It cannot say the page is good, or that it is yours.

So speed is no longer what you are paying for. Guidance is.

The proof: one real project

Netica Business Solutions is a strategic intelligence firm. Its website is five pages, in Spanish. Grafica built it.

The starting line. The first build took one day. It came from the firm’s sales presentation and used its brand’s colours, its two typefaces and the angled cut from its logo. Google’s test, run on 9 October, gives its home page 100 for speed and 100 for accessibility, on a phone and on a desktop.

That is where a capable build with today’s tools can arrive in a day. It is a finish line only if nobody asks the next question.

Looping screen recording of the first build of the Netica website, the starting line: its home page scrolling from the top in a desktop window 1440 pixels wide. It shows a headline over a photograph of a city at night, then a white section with a second headline and a photograph of an iceberg. — **Starting line.** Built in one day. The firm’s promise over a city photograph, then an iceberg and a strip of numbers.

Looping screen recording of the finished Netica website: its home page scrolling from the top in a desktop window 1440 pixels wide. It shows a headline over a dark wall of words, then a diagonal edge that wipes the dark screen into a white one as the page scrolls. — **Starting line.** Built in one day. The firm’s promise over a city photograph, then an iceberg and a strip of numbers.

The question no score asks. Correct is not the same as good. Does this page say what only this firm can say?

What guidance added

Two more days. This is what they went on.

A hard look. We read 25 current websites and laid ours beside them. Ours shared its parts with several: the cards, the number strips, the floating blocks.
A direction, written before the redesign. What the page argues, and the one motion that explains it.
Three ideas, built as working pages. The one chosen to lead is the one whose idea is the firm’s own argument: too much information, then a decision. The other two became inner pages.
The page turned around. The first build opened with the firm’s promise. The finished site opens with its client’s problem: “Nunca hubo tanta información. Tampoco tanta incertidumbre.” (There has never been so much information. Nor so much uncertainty.) In browsers that support it, as you scroll, a diagonal edge, cut at the angle of the firm’s logo, wipes night into day and the answer arrives. Wherever motion is switched off, the same two screens stand still.
Editing. Out went the photograph on the first screen, an iceberg and a lighthouse, four of the home page’s fourteen cards and all four on the services page. The finished site still has cards, number strips and stock photographs. They are no longer the idea of the page.
Then the experienced eye. A list of corrections of the kind no score reports: a word cut off on large screens, a logo that changed size from page to page, a word left alone on a line, a phone menu that was a word and is now three bars.

Looping screen recording of the first build of the Netica website, the starting line: its home page scrolling from the top in a phone-sized window 375 pixels wide. It shows a headline over a photograph of a city at night, then a white section with a second headline and a photograph of an iceberg. — **Starting line.** Home.

Looping screen recording of the finished Netica website: its home page scrolling from the top in a phone-sized window 375 pixels wide. It shows a headline over a dark wall of words, then a diagonal edge that wipes the dark screen into a white one as the page scrolls. — **Starting line.** Home.

Looping screen recording of the first build of the Netica website, the starting line: its services page scrolling from the top in a desktop window 1440 pixels wide. It shows a headline over a dark photograph, then a grid of four cards. — **Starting line.** A photograph, then four cards.

Looping screen recording of the finished Netica website: its services page scrolling from the top in a desktop window 1440 pixels wide. It shows a headline ending in one very large word, then four services set as text. — **Starting line.** A photograph, then four cards.

The numbers

Google’s own test on each home page, first load, before scrolling: three runs as a simulated phone and three as a desktop, on 9 October 2026. Middle values.

	Starting line	Finished site
Speed score, phone	100	98
Speed score, desktop	100	100
Main content on screen, phone	1.5 s	1.7 s
Main content on screen, desktop	0.34 s	0.32 s
Page weight, phone	190 KB	179 KB
Page weight, desktop	224 KB	168 KB
Files requested, phone	10	8
Files requested, desktop	10	7
Accessibility score, phone	100	100
Accessibility score, desktop	100	100

Re-run on 12 October 2026, three runs per version, middle values: phone speed score 100 for the starting line, 97 for the finished site.

Lighter. The whole site is less than half the size: 0.69 MB against 1.52 MB. On first load, before scrolling, the home page sends a quarter less data on a desktop (168 KB against 224 KB), about 5% less on a phone, and 7 files where there were 10.

Not free. Main content appears 0.2 seconds later on the test phone, the phone speed score reads 98 against 100, and on the phone the screen takes longer to fill: 3.8 seconds against 1.2.

Now read the table again. By score alone, the starting line wins: two points ahead on a phone, level everywhere else. Then watch the recordings. That is the gap a score cannot see, and it is the one your clients see first.

The race car

AI is the race car in all of this. It is why the first build took one day and the redesign two more.

But a race car still needs an experienced driver to run a good race. AI takes a website wherever it is pointed, quickly. Without the right guidance you still arrive fast. Somewhere else.

Before you hand anyone the keys

Ask who is driving, and what they questioned.
Ask to see what was taken out.
Ask for the numbers before and after, and which ones got worse.
Compare for yourself. The finished site: netica.solutions. The starting line, kept online on purpose: 152f2349.netica.pages.dev.

Building or rebuilding this year? Talk to us: grafica.group.

How we measured, and what we do not claim

Google PageSpeed Insights, both home pages in turn, first load, 9 October 2026, 12:02 to 12:14 p.m. Tegucigalpa time. Simulated tests, not real visitors; Google showed no real-visitor data for either address. 1 KB is 1,000 bytes.
Also slower on the finished site, and not in the table: the first thing on screen appeared at 1.53 seconds against 1.20 on the phone, and 0.32 against 0.26 on a desktop.
Google’s measure of how fast the screen fills: on the phone, a middle value of 3.8 seconds on the finished site against 1.2 (runs from 3.8 to 4.0, and from 0.9 to 2.3). On a desktop, 0.41 against 0.32.
Accessibility: our first measurement, on 7 October, read 98 on a phone and 94 on a desktop for the finished site, on two findings. Both were corrected on 8 October, and no accessibility check in Google’s test fails on either version today.
Whole-site sizes: every file of each version, fetched from its live address on the evening of 7 October, at full size, before the corrections of 8 October and not measured again: 37 files against 53. Cards were counted in the published files that night. The 25 websites, the corrections and the build days come from our dated working notes of 5 to 7 October.
One website, one client, three days; both versions are our own work. Not a controlled experiment, and nothing here is about visitors, leads or sales: the finished site went public on 7 October.
The first build was reviewed and corrected by the same experienced eye. What it had not yet had was the question and a written direction.
That the finished site is the better one is our judgment. The recordings are ours: both live sites on 9 October 2026, in the same browser, at the same window sizes (1440 and 375 pixels wide), each page scrolled the same distance on both sites, so you can make yours.