Agentic workflows are showing up everywhere, but most companies that want one don’t have the in-house expertise to build it. This is an early concept for a tool that lets a team assemble an agentic workflow without writing the orchestration code themselves.

The concept

The tool would walk a team through a short questionnaire about what they’re trying to automate, then let them build the workflow visually — picking from premade steps or defining custom ones, wiring them into a visual graph, similar to a DAG. A few ideas so far:

• A questionnaire that turns answers about the task into a starting workflow
• Testing as a first-class citizen, not something bolted on after the workflow works
• A visual builder for the graph itself, with premade and custom steps
• “Agent as computer” steps, similar to Operator mode in ChatGPT, where a step controls a UI directly instead of calling an API
• Prebuilt UI components for different entry points (chat, form, webhook, scheduled run)

Building AI-Powered Products has some good material on this that’s shaping how I’m thinking about it.

The canvas

The visual builder is the part I wanted to prototype first. Drag a step from the palette onto the canvas, connect steps by dragging from one port to another, then run a test pass to watch execution move through the graph in order.

Step details

Some visual thoughts on what each step needs under the hood. Nothing wires into the canvas above, but set the fields to see how each step’s config would read.