What we will never build, and why Birta starts where your agent stops

Most product pages describe what a thing does. That half sounds the same everywhere. What tells you where a product actually sits is the list of things it has decided not to become, written down before anyone asks so it can be held against us later.

So here is ours.

The short version is one sentence: your agent builds the site, and Birta is the layer that starts when your agent finishes. Everything below is a consequence of taking that sentence literally.

A non-goal is a decision, not a delay

There is a specific way to misread a list like this, and it is worth closing off first.

A backlog is a list of things not built yet. A non-goal is a list of things that would change the shape of the product if we built them, so we are not going to. From the outside the two look identical. They behave in opposite ways. A backlog item gets less true as the product grows. A non-goal gets more true.

The test we use is whether the thing would give the product a second job. A second job means a second set of decisions competing with the first, and the first one stops being done properly. Nothing on this page is an announcement, and nothing here is merely unfinished, because unfinished work belongs in a changelog, not in a list of promises.

We are not the first to write one of these down. Plausible keeps a page about when it is not the right fit, and the Bear Blog manifesto is essentially three sentences of refusal. Both are more useful than the feature lists next to them.

We will not build a website builder

No canvas. No drag-and-drop blocks. No visual editor where you nudge a heading three pixels to the left and press save.

This is the one people expect us to give in on, because it looks like an obvious addition rather than a different product. It is not a matter of taste. It is arithmetic about sources of truth.

A site that can be edited by your agent and by a canvas has two editors that do not know about each other. You move a section by hand, then ask the agent for a change to the copy, and the agent rewrites the file it last knew. Your manual move is gone. Or the canvas exports its own version of the page and flattens what the agent wrote. The usual escape from this is to make one of the two surfaces authoritative and the other decorative, at which point you have a builder with a chat box glued on, and the agent is no longer really building anything.

We would rather your agent be the only thing that writes your site. If you want the longer argument about where that leaves the established builders — including the cases where they are still the better answer — it is in WordPress vs Wix.

We will not build our own agent

You arrive with the tool you already use: a chat on your phone, a coding agent in your editor, a command-line assistant. We are not going to ship a Birta agent, a Birta chat window, or a Birta prompt box that competes with it.

The reason is that the tool people use changes faster than any of us could ship a replacement for it. Someone who was pasting into a chat window last year is in an editor now, and will be somewhere else next year. If publishing is tied to our agent, every one of those moves costs you your publishing setup. If it is tied to an open connection your tool already speaks, none of them do. That connection is a protocol rather than a product, which is worth understanding once if you have never looked at it.

There is a cost to standing here instead of one step earlier, and it is worth naming. We cannot fix a bad site. If the agent writes a weak headline or an awkward layout, that conversation happens in your tool, not with us. We hold the part after the files exist.

We will not run your application

Static sites and landing pages: finished HTML, CSS, browser-side JavaScript, images and fonts. No databases, no server-side code, no background jobs, no private application logic.

This one is already written up at length, because it is the boundary people meet first — the difference between a site and an application, and the same wall one layer up inside a chat artifact. The only thing to add here is that it is not a stage. It is the shape of the thing. A project that owns other people's data deserves a platform built to operate it, and the review that comes with one.

We are not the one holding your money

Birta has no checkout feature today, and this section is not a preview of one. It is about a role we do not want.

Being the party that holds your customers' money is a different company with a different centre of gravity: reserves, disputes, identity checks, the slow machinery that exists because other people's money is sitting somewhere it does not belong. If a site you publish ever takes payment, the shortest path runs from your customer to you. It should not detour through us, and the boundary is about that route, not about a date.

We will never ship a version you host yourself

Sites published through Birta run on our infrastructure. There will not be a download-it-and-run-it-yourself edition.

That is a dependency, and it would be dishonest to present it as anything else. Here is what we do instead of an escape hatch, and you can weigh it yourself:

• Your files stay yours. The agent that published them still has them and can write them anywhere else, without asking us for an export.
• The last ten versions stay available, and a bad change can be restored rather than rebuilt.
• The address is yours. Your own domain points at the site, and if you leave, it points somewhere else. Moving is a DNS change, not a support ticket.

What we are not offering is a copy of the platform to run in a corner of your own. Keeping one publishing path is what makes it possible to keep that path simple, and a second, self-operated one would quietly become the reason the first cannot change.

We will never promise you traffic

We do not create demand. A site does not generate attention by existing, and any product that implies otherwise is selling you the most expensive part of the job as a side effect of the cheapest.

Project analytics show you what happened to the attention you already had: how many people opened the site, day by day, and whether that number is moving. It is deliberately a small number rather than a dashboard, and it is genuinely useful. It is not a growth channel, and we are not going to describe it as one.

Three more, stated plainly

We will never sell you the parts. Not a marketplace of skills, templates, plugins and snippets to assemble yourself. The value is meant to be in the assembled path from a finished site to a live address. A shop full of components is a different business, and it is one that quietly moves the assembly work back onto you.

We will never become an operations system for a category. No bookings, no memberships, no class schedules, no staff rotas, no patient records. Every one of those is a real product with real depth, and none of them is a website with an extra tab.

We will never be an enterprise platform. No audit programmes, no contractual uptime guarantees, no procurement-grade security review. Those commitments are built into how a company operates, not bolted on, and pretending otherwise is how small products end up making promises they cannot keep.

The same list shaped this blog

This is not only a product habit, and the page you are reading is evidence.

Two shapes were considered for this site and rejected on purpose. The first was the standard website of our category: a marketing site with a feature tour and a modest blog, structured for a human visitor and nothing else. Proven, safe, and it treats the fact that an agent may be the first reader as an afterthought. We kept its marketing skeleton and dropped its hierarchy.

The second was a page farm — thousands of near-identical pages, one per keyword, each thin enough to be nearly free to produce. It works, in the narrow sense that it collects search traffic. It also reads exactly like what it is. We kept the idea of clusters and refused the thinness. We went and read thirteen competitor blogs to see what the thin version looks like from the outside, and the tells are not subtle.

And the part that usually goes unsaid: these articles are produced with AI. Deliberately, and by the same method we expect our users to work by. The line we care about is not AI against hands. It is whether a page contains something you could not have guessed before opening it — a real decision, a real limit, a real reason. That is the bar, and this page is trying to clear it.

When Birta is the wrong choice

None of those are failures. Two of them are people who have already solved this themselves, and two are people whose actual problem is a different one.

What is left when the list is done

A short list, on purpose. Everything here works today:

• Your agent publishes a finished site to a live address, and republishes it when you change your mind.
• It can edit files on the site through the same connection, without rebuilding and resending the whole thing.
• Images and fonts travel with the site. Video and PDF files still need somewhere else to live, so they do not travel with it.
• The last ten versions stay available, so a change that turns out wrong can be rolled back instead of rebuilt.
• Your own domain points at it, with HTTPS handled for you.
• Project analytics show how many people opened the site, day by day.
• An llms.txt file describes the site for the machines now doing part of the reading.
• Authorising your agent happens in a browser, so nothing depends on you pasting a key correctly.

That is the whole layer. It starts when your agent stops, it ends at your domain, and the list at the top of this page is what keeps it that size.

Frequently asked questions