Building Fiuto
In the last few months I independently built and designed Fiuto.ai, an end-to-end AI user research automation tool. To design and build it solo I had to rely heavily on agentic workflows, which I designed to replicate the way a real product team works, from discovery to design to delivery.
Fiuto's MCP server lets Claude Code and Cursor create and run studies from inside a build session. In the recorded sessions below, a collaboration feature went from the project in the tracker to production in one day, and a project opened by research reached staging in six hours.
Open fiuto.ai ↗Full design automation presentation ↗Start with part 01 →
The /ux family on a surface that already exists. One orchestrator and five skills in a single session: it grounds itself in the current stories, ranks what needs fixing, puts variants up for me to pick, translates the pick into components and checks what was built against it.
01Parts 01 to 02
Core flows in automated design
Redesigning surfaces that already exist: first a batch of simple admin components, then a complex panel with states and variants, both inside the coding agent and without a Figma handoff.
- Batch-updating simple components in an automated redesignHow I design and build shared admin components in one coding session, with no Figma handoff.
- Redesigning complex componentsHow I use a set of custom skills to redesign complex features inside the coding agent, without Figma.
One feature from question to production. The card on the board is remade at every step: grounded in research, narrowed to a direction, drawn as components, planned in phases, checked on staging, then live. The three marks on the track are the points where I decide.
02Parts 03 to 06
A new feature from design to delivery
One feature followed end to end: grounded from scratch, handed off as real components, planned and built in verified phases, then released to production.
- Designing new features with AIHow I ground a new feature in the code, the roadmap and the market before designing it.
- From an approved design to real componentsHow I turn an approved mockup into real components, with states, mobile and both themes.
- Planning and building a feature with the build agentHow I run planning, implementation and checks as one pipeline, and settle the decisions the agent hands back.
- Releasing the feature to productionHow I take a feature from staging to production.
Where a direction comes from. Four weekly scans feed one synthesis, and the synthesis reopens a roadmap decision: an option is struck out, the rest are ranked or parked. The two lanes on the right are the work that follows.
03Parts 07 to 09
Using research in automated design workflows
Where the direction comes from: a weekly competitive read that reopens a roadmap decision, the delivery that follows one of its findings, and the prototype a whole journey needs before it can be judged.
- Automated research for product direction and feature deliveryHow I turn a competitive finding into new projects on the roadmap in one session.
- From a research project to a delivered featureHow I take a research finding to a feature on staging.
- Prototyping a whole user journeyHow I prototype a whole user journey in the product's own components.
A QA pass run by agents. The orchestrator fans one build out to ten browsers, each driven by its own agent collecting passes and failures. What comes back is a single report: what held, what broke, and the defects ranked with the layer each belongs to.
04Parts 10 to 12
Reviewing and improving implemented experiences
What happens once the code lands: an implemented flow walked against the design it came from, the visual touches that give the product its tone, and a QA review run by agents in real browsers.
- Refining a feature after implementationHow I review a built onboarding flow against its prototype and send each fix back.
- Adding visual touches to working screensHow I place illustration and animation on a working screen in minutes, by asking the coding agent.
- Visual QA with a swarm of agentsHow I review a delivered feature with ten agents testing it in real browsers.