Create AI workflows for your business

Today we're introducing AI editing for automations. The editor's chat now turns a sentence into a drawn flow, built from your workspace, and nothing runs until you save it.

Where automations actually die

The flow your team described in a meeting rarely survives the builder.

Every operations team has the flow. It lives in a meeting note: greet the visitor, ask three questions, book the visit, alert sales. Everyone agrees it should exist by Friday.

Then someone opens the automation editor. The translation begins: which node, which branch, which setting on which tab. In a large company the person who knows the process is rarely the person who knows the builder. So flows queue behind the one colleague fluent in both. Many never get built at all.

The cost is every automation your team described and never shipped.

A chat that builds

The editor now has a chat that builds. You type, it draws.

Open any automation and the assistant is right there, in the panel you already use. Ask it for what you want. "Create a welcome that qualifies the visitor." The agent thinks for a moment. Then the canvas moves: a trigger, a question with choices, a form for contact details, a closing message, already wired in order.

Its reply lands as a drawn proposal, and the editor marks the flow as unsaved until you accept it. The chat edits as well as it creates. Ask for an email node after the trigger and it lands there, connected. Ask in your own language: the agent follows your panel.

It studies your workspace first

It designs the flow from your pieces, the ones already in your workspace.

Behind the chat there is an agent, the same kind that talks to your customers. When you send an instruction, it investigates before it draws. It checks the integrations your company has connected, the node types available, the documents in your knowledge base. Each step streams in the panel while it works, with the time it took.

That research shows in the result. A list of choices can point at your catalog instead of placeholder text. A branch can route through an integration you actually use. The proposal is assembled from your workspace.

Nothing saves itself

The AI proposes. The canvas shows it. You decide what survives.

Every reply from the agent is a proposal, drawn on the canvas in front of you and marked as unsaved. By itself, it touches nothing in your account.

You keep the tools an editor owes you. Undo rewinds your steps, the AI's included. A test panel runs the saved flow inside the editor before you enable it. For a company where automations touch revenue, that separation is the contract: the speed of a drafted flow, without the risk of a silent change.

What a workflow even is

A script your agent follows, drawn as boxes you can read.

Before the AI that builds them, meet the thing itself. A workflow is the script your agent follows in a conversation: each box is one step (a question, a list of choices, a form, a message) and each arrow is the path between them. No code hides behind it. What you see on the canvas is exactly what your customer will walk through.

And it runs. Watch the same flow execute: the trigger fires when a chat opens, each node lights up as it works, and the path the visitor took turns green behind them.

That readability is the point. When something misfires at step three, you see it at step three. When sales asks what the bot says after the form, you point at the box.

What an automation does for you

A flow guides the conversation, captures data and acts on it. On its own, every time.

We have never properly explained what our automations are, so here it is. In Hellomatik, an automation is a conversation flow that runs inside the chat your customers already use. It wakes up on a trigger, walks the visitor through a script you designed, and does real work along the way. No visitor waits for your office hours, and no two agents improvise different answers to the same question.

An automation does three jobs:

• It guides. Welcome menus, questions and choice lists steer the visitor to the right place: sales, support, a booking, a catalog. Conditions and switches branch the path by what the visitor answers.
• It captures. Forms collect several fields in one screen. A calendar node books an appointment with real slots. An upload node receives files. Every submission lands in a records grid, ready to export.
• It acts. The flow sends emails, calls a webhook on your systems, tags the contact, or generates a PDF from the data it just captured. When the script reaches its limit, an AI agent node takes over the conversation.

And it starts itself, on six triggers: the chat opens, a keyword appears, an intent is recognized, minutes pass in silence, a scheduled time arrives. The sixth matters most at scale. Another automation can call your flow as a subflow, so you build the booking step once and reuse it everywhere.

The full palette, node by node

Six families of blocks, and the AI composes from the same palette you use by hand:

Here is the part that matters: everything in this table is buildable by describing it. The agent picks from the same twenty nodes and six triggers you would. It wires choices to your documents, your database or another automation. And every text lives per language, up to six per canvas. If you can say it, it can draw it.

Five flows worth stealing

If you run a mid-size or large company, start with one of these.

Each one is a single sentence in the chat. The AI drafts it on the canvas; you adjust and save:

None of these requires a specialist. Each is a sentence in the editor's chat, then a review of the drawn proposal on the canvas.

What changes for a large company

The owner of the process becomes the author of the automation.

At enterprise volume, automation means a portfolio of flows across teams, languages and channels. The bottleneck was never the ideas: it was the queue behind the few people fluent in the builder.

The direction of travel is not subtle. Gartner sizes the market for software that enables this kind of automation at 1.07 trillion dollars by 2028, growing 13.9 percent a year. Back in 2021 the same firm predicted that pairing automation with redesigned processes would cut operational costs by 30 percent. McKinsey, after examining more than 2,000 work activities, calculated that about half of what people do at work is technically automatable with technology that already exists. The flows in this post are where that curve reaches your website.

The agent drafts and edits. It does not publish: enabling a flow is still a person clicking save, with the whole change in view. It is waiting in your automations editor today.