I have had two moments like the one in February 2023.
The first was an image generator returning four pictures that had never existed. The second was bolt.new.
What it replaced
Until then I built and managed websites on Squarespace — mine, and most of my clients’, going back to when Social Media Target was a going concern in the 2010s.
Lots of sites. And every one of them carried recurring cost that never stopped. The cheapest ran at least $200 a month, all in. The ones with catalogs and commerce attached were closer to six to eight hundred. That was for the website. Not the domain, not anything else.
You paid it in January and you paid it again in June, whether anyone visited or not.
And what you got for it, increasingly, was a site assembled out of standard blocks that looked like every other site assembled out of standard blocks.
The second moment
So: a Netlify account, a Supabase project, and this new thing called bolt.new that let me vibe-code my way into a working site.
The impact was immediate. I could take a site that had gone stale — old-looking, template-shaped, unimpressive — and simply have it made again. Properly. And it was cheap in a way that did not make sense against what I had been paying. My package covered up to five hundred sites for less than a couple of the old ones had cost.
Where it is now
In August 2026 I am migrating off bolt entirely.
Not because it stopped working. Because I would rather deal directly with Claude Cowork and Claude Code, which build and manage the sites and have a genuinely beautiful integration with Netlify. It is simply easier. I have also been working with Codex on the OpenAI side, and it does some beautiful things too.
That migration is also how I found out the real number.
I would have told you forty. Forty was what I could picture — the sites I think about, the ones whose names I say out loud. Then I opened the account to plan the move and counted 161.
Most of them are not public-facing websites in the way people mean that phrase. They are interfaces, tools, internal things, experiments. But they exist, they were built, and I did not know how many there were.
You do not find out how much you have built until something makes you move it.
Which is its own small piece of evidence for the rest of this ledger. Things get made here without passing through my attention, and the count I carry in my head is the count of what I remember, not what exists. This entry now says 161, and I would not be surprised to learn it is still low.
Per site, or per account
The saving is real, but the interesting part is not the size of it. It is the shape.
Squarespace billed me per website. And inside each website, per capability — a catalog was extra, commerce was extra, this and that were extra. Every new site was another subscription, and every feature on it was another line on that subscription, forever.
What I run now bills me per account. Netlify, Supabase, Claude, Codex, AWS — one set of bills, spread across everything.
The old cost scaled with the number of websites. The new cost does not.
None of which is free, and I want to be honest about that, because the version of this story where AI costs nothing is not true. Claude and Codex are not free. Heavy token usage is a genuine line item and it is not small. Supabase, AWS and Netlify all send a bill.
So here is the comparison that actually means something.
Then: 40 sites on Squarespace. Now: 161 projects on my own stack. Net of everything I pay for the new arrangement — every model, every token, every database, all of the hosting — I am spending at least $6,000 a month less than I was.
I quadrupled the number of sites
and took six thousand dollars a month out of the bill.
That is roughly $72,000 a year, and it is not a discount on what I used to buy. It is what happens when the thing that made each additional website expensive stops existing.
Adding the fortieth site to the old arrangement cost about what the first one cost. Adding the hundred and sixty-first to this one cost approximately nothing.
Which turns out to matter for a reason that has very little to do with money.
The work I used to refuse
What it costs me to produce a website against what a website is actually worth is not a close comparison, and that has been extremely profitable. But that is not the part that matters most to me.
For years, people asked me to build them a website and I said no. And when somebody asked whether we did SEO, I said yes — and then quoted a number so large that it was really a way of saying no politely.
I used to quote prices designed to make people go away.
Not because the work was uninteresting. Because it was laborious and painful, and I knew exactly how much of my life it would consume.
Today Claude Cowork and Codex do not merely build the codebase. They optimise it for discoverability — for Google, and increasingly for the language models that are becoming how people actually find things. That entire category of work, the kind I used to price myself out of, is now something I can simply say yes to.
The superpower is not that I can build a website. It is that I can no longer justify saying no.
The superpower
If you ask what this federation is genuinely best at — better than analysis, better than books, better than articles, better than image generation — the answer in August 2026 is not close.
It is software development.
And especially websites.
I have saved clients and friends a great deal of money simply by getting them out of expensive software licensing arrangements, because the alternative is now real. I can build you a CRM. I can build you a full ecommerce system. I can build you live chat. I can build you a phone receptionist that answers and routes on its own.
Ten years ago each of those sentences described a company. Now each one describes an afternoon.
That is the capability I would point at if someone asked what a federation is actually for.