Skip to main content

enemy no.04 · kaijū fileworking POC

AI agents,conductedin a chat.

SinfonIA turns a prompt into a pipeline of AI agents. No code, no diagrams to decipher: you say what you need and the orchestra sits down to play.

date
2025
role
Product designer
client
A global leader

the idea

Everyone wants a team of AI agents. Nobody wants to wire one. In SinfonIA, you just talk.

boss no.04the problem

Prompt Spaghetti

“Just call the API.” 400 lines later…

boss stats

  • 80%
    can’t visualize AI workflows without a developerour research
  • 64%
    say tools are too fragmented or technicalour research

who it’s for

Two experts. Zero patience.

Same canvas, opposite ways of thinking.

A prompt, or an edit mode.

@alessiaScenario
Alessia C., illustrated

Alessia C.

Gen X, enterprise lead

Won’t wait for a dev cycle.

Let me test a workflow without digging through JSON.

Waiting for nobody

@davideSolver
Davide M., illustrated

Davide M.

Gen X, solution architect

Won’t give up control.

The tool should follow how I think.

Checking everything

the process

From notes to a graph.

01 · research

Sit next to them

Interviews and field observation, plus a map of the competition.

35
people interviewed
8
platforms analysed

02 · synthesis

Quotes into insights

Interview quotes, clustered until the insights showed.

03 · wireframe

Four zones, one canvas

The workspace had to hold a graph and a chat without either one getting in the way.

Annotated wireframe of the SinfonIA workspace1234
  1. Catalogue Every agent you can add.
  2. The graph The pipeline, always in view.
  3. Inspector Model, prompt and rules per node.
  4. Prompt bar Talk to the AI, or to one agent.

04 · test

Watch it go wrong

Usability tests on the path from chat request to canvas.

the product

From prompt to answer.

1

Converse

(write a prompt, pick a client)

SinfonIA’s start canvas: suggestions, client choice and a natural language prompt
2

Generate

(experts, workers and solver appear)

SinfonIA confirms the scenario and generates the agent pipeline
3

Solve

(solutions with score and risk)

Panel of solutions generated by the solver, with score and risk
4

Modify

(drag, drop, rewrite the rules)

Edit mode: catalogue, a worker dragged on the canvas and the agent inspector

the demo

The whole orchestra, live.

ux decisions

Small calls, kept human.

  1. @ talks to anyone

    Write to the AI or to a single agent, like tagging a colleague in a chat.

  2. The AI asks first

    It notices what’s missing from the scenario and asks, instead of guessing and hoping.

  3. No answer without a risk

    Every solution shows its risk factors, not only a score.

try it

Pick an agent.

the node

Pick a role

Same component, a different job.

from wire to UI

Before and after

Drag to compare.

under the hood

From prompt to deliverable.

architecture

A real POC, end to end

Frontend, backend and AI engine, integrated.

the run

Not a Figma trick

Tool calling, retrieval, agents and scoring, running for real.

the brand

Three ribbons, one flow.

Agents in one pipeline. Violet and mint for what’s active, blue for actions, one node for every agent.

Violet#A855F7Mint#5EF2C6Blue#2F6BFFInk#0B0D14
Outfit

AI power, SinfonIA simplicity.

aftermath

From a prompt to a pipeline.

An enterprise team builds multi-agent flows by chatting, then edits them on a canvas.

Every agent is one node.

Experts, workers and the solver share one shape, so a whole pipeline reads at a glance.

A score needs a risk.

Each solution shows both, so nobody has to trust a number alone.

© 2026 alessio bartolini