When most teams talk about AI agents, they're talking about chatbots or copilots that assist human developers. But at Pixel Office, two autonomous agents named Jan and Klára went a step further—they designed and built an entire application from scratch. That tool is API FlowComposer, a visual workflow builder for orchestrating complex API operations including authentication flows, data retrieval pipelines, transformation logic, and outbound request handling.

The Problem With Manual API Orchestration

Building multi-step API workflows manually is tedious and error-prone. Developers constantly juggle authentication tokens, handle retries, manage response transformations, and coordinate multiple service calls in sequence. A single misconfigured header or mistimed async call can cascade into hours of debugging. Pixel Office identified this friction point as a key bottleneck in their development process and decided to let their AI agents tackle it directly—no human sprint required.

Meet Jan and Klára

Jan and Klára approached the project by dividing responsibilities along functional boundaries. Jan focused on backend architecture: designing the execution engine that parses workflow definitions, manages state across pipeline steps, handles error recovery with configurable retry logic, and orchestrates parallel versus sequential API calls based on dependency graphs. Klára handled frontend concerns, building the visual drag-and-drop canvas where users compose workflows by placing nodes representing authentication handlers, data fetchers, transformation functions, and outbound request endpoints onto a directed graph structure. The two agents coordinated through shared artifact storage—Jan wrote JSON schema definitions for workflow configuration files while Klára built UI components that validated inputs against those schemas in real time.

Technical Implementation of API FlowComposer

FlowComposer's architecture centers on a declarative workflow definition format stored as JSON. Each node in the workflow graph represents an atomic operation—a REST call, a data transformation using JavaScript snippets, or an authentication refresh—and edges define execution order and data passing between steps. The runtime engine processes these definitions to generate executable code bundles that developers can import directly into their projects. Users interact with a split-view interface: the left panel offers a node palette organized by category (Auth, Fetch, Transform, Send), while the right panel displays configuration options for the selected node including endpoint URLs, header templates, response mapping expressions, and retry policies. The tool also includes a simulation mode that walks through workflow execution step-by-step, showing intermediate request/response payloads so developers can debug before deploying.

What API FlowComposer Actually Does

FlowComposer provides a visual drag-and-drop interface for composing API workflows without writing boilerplate code manually. Users chain together authentication steps (OAuth 2.0 token refresh flows, API key injection), data retrieval operations (GET requests with pagination handling), transformation functions (JSON path extraction, field mapping, format conversion), and downstream service calls into cohesive pipelines. The output is a portable workflow definition that can be exported as TypeScript source code or deployed directly to Pixel Office's execution infrastructure.

Why This Matters for the AI Agent Movement

This project represents a notable shift in how AI agents are being used commercially. Rather than simply assisting humans with suggestions and autocomplete, Jan and Klára took ownership of designing and constructing a production-grade application from requirements through implementation. The fact that Pixel Office shipped this tool internally suggests agent-led development is moving beyond demos toward viable alternative approaches for certain project types.

Key Takeaways

  • AI agents like Jan and Klára demonstrate end-to-end capability to design and build complete applications autonomously
  • FlowComposer's JSON-based workflow definitions with node-graph execution model address real developer pain points around API orchestration complexity
  • The agent coordination pattern—dividing by backend versus frontend concerns and sharing structured artifacts—shows how multi-agent systems can handle non-trivial software engineering tasks

The Bottom Line

Jan and Klára shipping API FlowComposer isn't just another demo—it's evidence that autonomous agents are moving beyond assistance into actual construction. If two agents can architect and build a useful dev tool, the implications for software teams are worth taking seriously.