Introducing Multi-Agent Editing in the Stack48 Editor

Some changes to a website are quick: a new phone number, a fresh photo, a tweak to a headline. Others are much bigger. "Add a pricing page, an FAQ page and a team page" is one sentence to type, but it's a lot of work for whoever is building the site.

The Stack48 editor now handles those bigger requests differently. When you ask for a substantial change, the agent that edits your site can bring in helper agents to share the work. You still type one request. You still get one result and one undo. The difference is what happens in between.

What was slowing big requests down

Until now, every request in the Stack48 editor was handled by a single agent, working through the job one step at a time. For small edits that's perfect. For a request that touches several new pages at once, it meant waiting while each page was built in turn, and on the largest requests the agent could run out of time before it finished.

Multi-agent editing is built to fix exactly that.

How multi-agent editing works

When you send a request in the Stack48 editor, the agent that picks it up acts as the lead. It reads what you've asked for and decides whether helpers would get the job done better. You don't need to switch anything on or configure anything. If a request is small, the lead simply does it.

There are two kinds of helper.

Explorers: quick answers about your site

Explorers are read-only helpers. The lead sends one out when it needs to know something about your site before making a change, for example "where is the contact form handled?". The explorer looks, comes back with a short summary, and never changes a single file.

Explorers are available on every plan, including Free.

Writers: building separate parts at the same time

Writers do the building. When a request naturally splits into two or more substantial, independent pieces, such as new pages or large new sections, the lead hands each piece to its own writer. Up to three writers work in parallel, each looking after its own part of the site.

The lead keeps hold of anything shared and does it first. Your navigation, layout, shared components and the structure of your site's data all stay with the lead, so every new page fits the rest of the site rather than going its own way.

Not every request is worth splitting up, though. If you ask for a small change repeated across lots of existing pages, say updating the same line of text on eight pages, the lead does it in one go itself. In our testing that was faster and cheaper than handing it out, so that's what the editor does.

One request, one result, one undo

However many helpers join in, you get a single, tidy outcome:

• Merged and checked once. When the writers finish, their work is brought together, checked once and saved as one change.
• Undo works as before. One click takes the whole change back, exactly as it always has.
• Overlaps are handled for you. If one writer's work would clash with something else, only that piece is dropped. Everything else is kept.
• Nothing is left half-done. If a helper runs out of time, the lead picks up that piece and finishes it.
• Stop means stop. Pressing Stop halts every helper straight away.

See every task as it happens

While helpers are working, a live Working in parallel card appears in the editor. It shows one row for each task, with:

• its status
• the step it's on right now
• the files it has changed
• how long it has taken

Expand any row to see that task's activity in more detail. When everything is done, the final message groups the changes by task, without repeating itself, so it's easy to see what changed and where.

If something doesn't go to plan, the editor tells you plainly. You'll see messages such as "Couldn't finish this step", "Ran out of time for this step" or "These changes overlapped with other work and were not kept", rather than anything alarming.

What's included on each plan

• Free: up to 3 writers per request (the same as Pro, for now), plus explorers
• Pro: up to 3 writers per request, plus explorers
• Agency: up to 5 writers per request, plus explorers

You can compare everything else that comes with each plan on our pricing page.

Does it cost more?

No. Pricing hasn't changed. You pay for the actual usage of every agent involved in a request, at the same rate as before.

In fact, splitting the work can help. In our testing, adding two new pages with writers used noticeably fewer credits than having a single agent build both on its own.

A few other things worth knowing:

• Low balance? It still works. If your balance is running low, the editor quietly does the job with a single agent instead. You won't see any billing message.
• Stop and revert costs nothing. If you stop a request and revert it, you aren't charged.
• Stop and keep pays for what was done. If you stop but keep the work so far, you pay only for the usage up to that point.

What this means for you

If you're building a new site or reshaping an existing one, you can ask for the bigger things in one go. "Add a services page, a gallery and a contact page" or "create a page for each of our three locations" are now the kind of requests the editor is designed to split up and build side by side, while keeping your navigation and layout consistent.

If you're new to Stack48, how it works explains how a site goes from a short description to a finished, editable website, and our page on multi-page websites covers how pages fit together.

Multi-agent editing is live now in the Stack48 editor. Open your site, ask for something big, and watch the Working in parallel card do its thing.