Spec-Driven Development with ATDD & AI (Valentina Jemuović)

You write a spec, hand it to AI, and it gives you back a PR that looks right. But the implementation can be wrong!

In the current tools (Kiro, spec-kit, Tessl), the spec is a Markdown file AI is supposed to follow. But AI can ignore the spec or misread it.

So you end up testing everything by hand, and QA keeps sending the PR back because it doesn't work.

In this session, you'll see how to use ATDD to practice spec-driven development with executable specs. Starting from the acceptance criteria, AI writes failing acceptance tests (RED), and you review and approve them. Then AI implements the system code until the acceptance tests pass (GREEN), which shows that the acceptance criteria are met.
We'll illustrate this with an eShop example, so you can see what spec-driven development with ATDD looks like in practice.

What we'll cover:

- 𝗩𝗶𝗯𝗲 𝗰𝗼𝗱𝗶𝗻𝗴 & 𝘀𝗽𝗲𝗰-𝗱𝗿𝗶𝘃𝗲𝗻 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁 - Where is your team today?
- 𝗘𝘅𝗲𝗰𝘂𝘁𝗮𝗯𝗹𝗲 𝘀𝗽𝗲𝗰𝘀 - Why a Markdown spec isn't enough?
- 𝗧𝗵𝗲 𝗥𝗲𝗱 𝘀𝘁𝗲𝗽 - AI writes the Acceptance Tests, DSL & Drivers
- 𝗧𝗵𝗲 𝗚𝗿𝗲𝗲𝗻 𝘀𝘁𝗲𝗽 - AI implements code to make the Acceptance Tests pass
- 𝗔𝗜 𝘂𝗻𝗱𝗲𝗿 𝗰𝗼𝗻𝘁𝗿𝗼𝗹 - Control scope of changes & optimize token costs
Live Q&A at the end.

ABOUT VALENTINA

Valentina Jemuović is a Technical Coach at Optivem. She helps engineering teams accelerate delivery through Spec-Driven Development with ATDD/TDD & AI, so they can release frequently without fear of regression bugs. She has coached teams across Europe and North America in finance, logistics, and industrial automation to accelerate delivery & reduce bugs.

- LinkedIn: https://www.linkedin.com/in/valentinajemuovic/
- Twitter: https://x.com/valentinajemuov
- Github: https://github.com/valentinajemuovic
- Substack: https://journal.optivem.com/
- Optivem: https://optivem.com/