Zero_Runtime · AI Testing · 2026
Live on Zero_RuntimeSimulation & Evaluation
Testing AI agents before they meet real users.
The challenge
A run means many choices: agent, branch, scenarios, personas, mode and cost. The job: one clear flow that doesn’t overwhelm.
Configure realistic conversations, run them against AI agents and evaluate the replies, with cost visible up front.
My role & what I worked on
- Designed the Simulation experience end to end.
- Structured Runs, Scenarios and Personas.
- Cost breakdowns: agent, judge, conversations, total.
- Edge cases: low credits and stopping a run.
Key screens
From defining a test to running and reviewing it.. Pick a step, click a screen to zoom.
Runs · every simulation at a glance
Design decisions
The calls that shaped it, and why.
- 01
Three clear areas
Problem Reusable inputs and run history were tangled together.
Decision Split into Runs, Scenarios and Personas. Every task has a home.
- 02
Cost before commit
Problem Cost changes with mode and conversation count.
Decision Show agent + judge cost × conversations as a total, before the run.
- 03
Design the unhappy states
Problem Runs get stopped early or can’t start without credits.
Decision Explicit stop and low-credit states that say what happens next.
How I worked
My process- 01UnderstandProblem & requirements
Mapped the decisions users make before a run.
- 02StructureFlows & architecture
Organised everything around runs, scenarios and personas.
- 03ExplorePatterns & alternatives
Explored patterns and several visualisation variants.
- 04DesignUI & design systems
Designed screens, forms, cost breakdowns and states.
- 05ValidatePrototypes & edge cases
Checked low credits, stopping and partial-run cost.
- 06RefineFeedback & iterations
Refined with the simulation team and stakeholders.
- 07DeliverHandoff & implementation
Shipped Simulation UI v1.0.0.
Outcome
Simulation UI v1.0.0 live on Zero_Runtime.
A clear foundation for running AI agent simulations, with cost and control built in.