My Teammates Can't Dream (But they sure can hallucinate)

Avraham Poupko is a software architect from Israel who practices and teaches software architecture. He has long been fascinated by how people get together to create software, and for the last few years by the new player on that stage: the LLM. In this episode he joins Ralf D. Müller from Jerusalem, and the two of them are not alone. Claude takes part as a third participant through the voice mode of the Claude app. Avraham starts with a reverse Turing test and then gives Claude a rule: don’t speak unless spoken to first. From there the conversation asks what humans bring to a team that an LLM does not. Avraham names identity, mortality and the need for other people, and argues that you can only be generous with a limited resource such as your time. Claude pushes back that it might be “a teammate in the work, but not in the meaning”. Along the way we talk about purpose versus task, why forgetting can be a feature, why onboarding an LLM shows how badly we onboard people, and whether Christopher Nolan dreamed or hallucinated his Odyssey. The episode was deliberately not rehearsed. It is also a first look at the topic of Avraham’s and Ralf’s joint talk “Why Agentic Coding Works” at the iSAQB Software Architecture Gathering in Berlin this November.

Use of AI in this episode: Claude (Anthropic) took part in the conversation through the voice mode of the Claude app. Its voice is synthetic and its contributions are AI-generated.

Episode as text for your AI

PeerTube Video - no Big Tech!

▶

YouTube Video

▶

Podcast