Spec-Driven Development with AI Unified Process & AI Agents

Many software projects are still code-centric: requirements are loosely defined, code is written, and documentation, tests and specifications fall behind. The result? Specifications become outdated, documentation is incomplete, and tests are missing.

In this workshop, you explore a different approach: Spec-Driven Development. Using the AI Unified Process and hands-on labs with an AI agent, you learn how to write a clear, Markdown-based specification that stays in the repository and evolves together with the code (spec-anchored). The AI agent generates code and tests from it, and you check the result against the specification.

You also learn how to build a harness around the agent so that the results stay reliable, how to apply the approach to existing systems, and what works in real customer projects.

The Workshop

Vaadin1

The Goals

  • You understand the difference between code-centered and spec-centered work.

  • You can choose a suitable method for your project: OpenSpec, BMAD, GitHub Spec Kit, Kiro or the AI Unified Process.

  • You know the AI Unified Process and can write use cases that every stakeholder can read.

  • You have experienced how an AI agent generates code and tests from specifications.

  • You know how to build a harness with guides and sensors that makes the agent reliable.

  • You know how to use Spec-Driven Development with existing systems and in teams, and how to keep risks and token costs under control.

The Agenda

  • Spec-Driven Development: Why?

    • Problems of code-centered work: specifications become outdated, documentation lags, tests are missing.
    • The three levels of Spec-Driven Development and why the specification should stay in the repository.
    • AI is not a compiler: steps of AI adoption and the move from developer to builder.
  • Choosing a Method

    • OpenSpec, BMAD, GitHub Spec Kit, Kiro and the AI Unified Process at a glance.
    • Which method fits when: greenfield or brownfield, lifetime of the system, regulation, team size.
  • AI Unified Process and Use Cases

    • From vision and requirements to use cases, entity model and architecture, then code, tests and delivery.
    • Use cases and user stories, levels of detail, behavior instead of code.
    • One specification for the whole lifecycle: estimate, implement, test, operate and change.
  • The Agent

    • AI agents with skills, MCP and subagents. The approach works with any AI agent.
    • Security model, sandboxing and real isolation.
  • Hands-on: Hotel Reservation System

    • From the vision to requirements, entity model, use case diagram and use case specifications.
    • Review the specification and get it ready for implementation.
    • Generate database migrations, code and tests from the specification.
    • Check the coverage of the specification and change a requirement.
    • Bonus: model the business process in BPMN and generate end-to-end tests from it.
    • Goal: by the end of the day, a small app is running, fully developed from the specification.
  • Harness Engineering

    • Guides and sensors: AGENTS.md, skills, hooks, ArchUnit and quality gates.
    • Architecture with the 4+1 view model, ADRs and enterprise constraints as input.
  • Brownfield Projects

    • Reverse engineer the specification from existing code and change the system from the spec.
  • Lessons from Customer Projects

    • Impact on architecture and teams: modular systems, self-contained systems, review as the new bottleneck.
    • Token economy: where the cost comes from and how to reduce it.
    • Risks, mitigation strategies and how to try it on your own stack.

The Target Audience

  • Software developers, architects, business analysts, requirements engineers, product owners, test engineers and team leads interested in AI-driven software development

The Requirements

  • Basic programming knowledge

  • A laptop with an AI agent of your choice, or a GitHub account to use GitHub Codespaces (nothing to install)

  • For the optional second day: your own code base and permission to use it with an AI agent

Duration and Format

  • 1 day: foundations, AI Unified Process, harness engineering and hands-on labs with the Hotel Reservation System

  • Optional second day: hands-on with your own code. Pick a real use case from your project, write the specification or reverse engineer it from the existing code, build a minimal harness for your stack and let the agent implement and test it. We review the results together and measure the token cost per use case.

  • Online or on-site

Next Public Workshop

  • Tuesday, 8 December 2026, 14:00 to 17:00 CET, online (Zoom)
    A compact 3-hour hands-on workshop: write a system use case, generate an application from it and change the features through the specification.

  • Standard ticket USD 390, early bird USD 290 until 25 November 2026

The Books

Customer Voices

Feeback Orange
Workshop: Spec-Driven Development with AI Unified Process & Claude Code

As Head of Software Engineering at Orange Business Digital Switzerland AG, I always enjoy working with Simon Martinelli as a trainer for our in-house software engineering training courses.
His workshop on Spec Driven Software Development with AI agents was also excellently prepared, practical and tailored precisely to our needs. In my opinion, its underlying approach (AIUP) is very convincing and clearly shows how modern software development with AI can develop sustainably.
I particularly appreciate Simon's eye for the big picture and his ability to convey complex topics in an understandable and tangible way. Our developers were able to pick up a lot of concrete information and apply it directly. I can fully recommend Simon as a trainer and would be happy to book him for our teams again.

SimonSpeaker

Thomas Wenger | Head of Software Engineering
Orange Business Digital Switzerland AG

OrangeBusiness_Feedback
Feedback Orange
Workshop: Spec-Driven Development with AI Unified Process & Claude Code

As Head of Software Engineering at Orange Business Digital Switzerland AG, I always enjoy working with Simon Martinelli as a trainer for our in-house software engineering training courses.
His workshop on Spec Driven Software Development with AI agents was also excellently prepared, practical and tailored precisely to our needs. In my opinion, its underlying approach (AIUP) is very convincing and clearly shows how modern software development with AI can develop sustainably.
I particularly appreciate Simon's eye for the big picture and his ability to convey complex topics in an understandable and tangible way. Our developers were able to pick up a lot of concrete information and apply it directly. I can fully recommend Simon as a trainer and would be happy to book him for our teams again.

Thomas Wenger
Head of Software Engineering
Orange Business Digital Switzerland AG

OrangeBusiness_Feedback
SimonSpeaker

My IT courses are fully customizable to your needs. I offer training both online and on-site – ideal for teams of 5 to 15 people.
I would be happy to create a tailored offer that fits your requirements!