ActionAI · Enterprise AI automation · Feb 2025–present
From hand-coded AI workflows to a platform engineers can scale
ActionAI’s AI engineers were building enterprise workflows by hand, in code. I joined at POC stage, when only a few pieces were standing and the UI was barely there, to turn that practice into a product: a platform where they build, run, evaluate and fix workflows until the reliability number is good enough to ship.
Lead designer. Every major product decision, together with the CPO and the Head of Design. Hands-on design of the core features; built the design system from scratch; helped and reviewed the three designers who joined later.
The AI engineering and R&D teams, daily. The product’s flows came from research with ActionAI’s own AI engineers.
Where it started
Week one was getting everyone synced on the technicalities: the team held different ideas of what the platform’s core was. Only then could the POC become a product.
What I did
01
Made the product a loop
Build → Run → Evaluate → find the bottleneck → back to Build. Every screen hands an engineer to the next step with the evidence they need, so iterating toward a high reliability score is fast.
02
Untangled the complex blocks
Data formats, loops and logic were the hard part of moving from code to canvas. I worked closely with the engineering and AI teams to turn those into nodes an engineer can read and modify with UI or code.
03
Reliability as one number
Large sample runs collapse into a reliability score with the issues ranked under it. Bottleneck analysis drills a subset of samples to the root cause and jumps back to the node that caused it.
04
One flow from design to build
Design and handoff moved into the same flow as the code. The design system lived in the repo, and designs reached engineers as pull requests, not Figma links. I brought design and development onto that flow: handoffs got cleaner, design and front-end moved faster, and more iteration happened before handoff.
POC → production
Enterprise clients in legal, commerce, logistics, finance and the public sector.
ActionAI publishes 18k+ hours saved and 15× efficiency gains across its clients.
Those are the product’s numbers. My part was the loop they run on.