From idea to finished code in a single loop

Spexus moves your idea through four stages: specification, planning, execution, and acceptance. A specialized AI agent supports each stage.

Request demo Back to home

How it works

You describe the idea in your own words. Spexus turns it into working code through a chain of four AI agents, each responsible for its own stage.

Step 1

Stage 1: Analyst

From a raw idea to precise specifications

You describe what you want to build in plain language. The AI analyst asks follow-up questions and helps turn the idea into a full document set.

1
  • Epic - project goal, term glossary, and C4 diagrams
  • User Stories - concrete scenarios of who does what and why
  • Requirements - detailed EARS-aligned specifications without ambiguity
  • Acceptance Criteria - explicit conditions in Given/When/Then format
  • Architecture - components, interfaces, data models, and technology choices
Input

Input - "I want to add PDF export for reports."

Output

Output - a complete specification ready for implementation.

Step 2

Stage 2: Planner

Specifications are broken down into AI-ready tasks

Once the specifications are ready, Spexus decomposes them into tasks, a set of concrete prompts for AI coding agents. Each task implements one slice of the functionality.

2
  • Tasks are ordered correctly with dependencies in mind
  • Each task contains everything AI needs to implement it: what to build, which requirements to respect, and which standards to follow
  • Task size is tuned so an AI agent can complete it in a single iteration
Input

Input - a specification with 20 requirements.

Output

Output - 15 tasks ready for execution.

Step 3

Stage 3: Executor

An AI agent writes code from the tasks

The execution agent takes tasks from the plan and completes them one by one. For each task it receives the full context through MCP.

3
  • Requirements and acceptance criteria from the specification
  • Steering documents - architecture, code style, and team standards
  • Outputs of previous tasks so the code stays consistent
  • Supported AI agents: Claude Code, Cursor, OpenAI Codex, Windsurf, Cline
Input

Input - a task list with context.

Output

Output - implemented code.

Step 4

Stage 4: Acceptor

AI verifies the code against the requirements

The acceptance agent compares the implementation with acceptance criteria from the specification and returns the result in a clear report.

4
  • Walks through every Given/When/Then scenario
  • Checks that the code implements exactly what the requirements describe
  • Builds a report showing what passed, what failed, and where the gaps are
  • If something is off, the task goes back to the executor for rework
Input

Input - code and acceptance criteria.

Output

Output - validated code aligned with the specification.

Summary

You arrive with an idea. You leave with code.

Each stage is supported by a specialized AI agent. You keep control of the process and the decisions, while AI handles the execution.

Before After
"I need PDF export" Specification, code, and tests ready for deployment
A team of 10 people for 2 sprints One person + Spexus in days
Context gets lost between stages Every agent sees the full picture

See this on a real use case

Leave a request and we will show how the four-agent flow works on your product.

Request sent. We will show the AI agent flow on your use case.