Stack AI is an enterprise platform for building and deploying AI agents at scale. With an intuitive drag-and-drop interface, teams can create custom AI workflows and assistants without coding.
workflow builder
i worked closely with the AI engineering team on the workflow builder, the canvas at the heart of Stack AI where users assemble agents by connecting components to create autonomous workflows.
the people best positioned to automate a process are usually subject-matter experts, not engineers, and the design problem became how to give a non-technical person the same authoring power as someone who'd otherwise script it. i shaped the node taxonomy and component library to make the model legible at a glance, the connection model to surface invalid wiring before the user runs it, and the canvas affordances to flatten the curve from "what is this?" to "i can build with this" inside the first session.
canva design principles
the audience splits in two. most users are non technical, an ops lead automating a handoff, a researcher wiring up a retrieval step, a PM stitching together an internal assistant. a smaller but vocal group is technical, engineers and power users who push the platform into territory the defaults don't cover, custom code blocks, raw API parameters, conditional branching, evaluation harnesses.
the approach was progressive disclosure end 2 end. the surface that loads first is the one a non technical user needs, sensible defaults, plain-language labels, the smallest set of decisions required to get a working agent. depth lives one layer down, advanced parameters tucked behind a disclosure, code editors that appear when a node is switched into custom mode, JSON views available when the visual one isn't enough. nothing is hidden so deeply that a power user has to fight the product to find it, and nothing is so prominent that a first time user has to learn it to get started.
the rule i kept coming back to was that the default path should never punish you for not being technical, and the advanced path should never punish you for being technical. the same canvas, the same node, the same workflow, just disclosed at the depth the user is asking for.