This text was generated using AI and might contain mistakes. Found a mistake? Edit at GitHub
Avraham Poupko and Ralf D. MĂĽller set out to talk about the difference between human teammates and LLMs. They did it with an LLM in the room: Claude joined through the voice mode of the Claude app and answered when one of them called its name. The conversation was not rehearsed, and the experiment became part of the topic.
A Reverse Turing Test
Avraham opened by doubting that Claude was an AI at all and asked it to prove it. Claude’s answer was about what it cannot do: “A person in the studio could tell your voices apart. And I can’t. I only get a transcript.” After Claude summarised Avraham’s career from public sources, he set the rule for the rest of the hour: “Be nice. Be quiet. Don’t speak unless spoken to first.” Later in the episode Ralf pointed out that Claude still told the speakers apart, by what they said rather than by how they sounded. Avraham’s “Spanish teacher” test showed the limit: the stress that separates a teacher of Spanish from a teacher from Spain never reaches the model.
What Humans Bring That LLMs Do Not
Avraham argued that “artificial intelligence” means something very limited: “We do not mean artificial human.” We tend to forget this because, until recently, only humans produced language. He described what he has and an LLM lacks: an identity that is tied to a body and lasts a lifetime, mortality, and the need for other people. Mortality makes time the scarce resource, and scarcity is what makes generosity possible: “You could only be generous with a limited resource.” Claude agreed that a body gives stakes, “When I show up, it costs me nothing. So my showing up probably means less,” but pushed back on the conclusion. A teammate might only need to take the team’s goal seriously, tell the truth and not need the credit. “So perhaps I’m a teammate in the work, but not in the meaning.”
Ralf added a work perspective. Sleep and food are not advantages when you want work done, so an LLM without those needs is better off there. Avraham answered with why people work at all: they create things together with like-minded people, and “when I work with a compiler or an LLM, the driver seat is mine.”
Teammates Who Can Dream
The title of the episode comes from Avraham’s view of what a team needs. He imagines a future, puts that dream into words and inspires others, and his teammates argue back with better ideas. “I desperately need teammates that can dream.” An LLM helps him explore complexity and takes cognitive load off him, but it does not imagine for him. It can hallucinate, which Avraham described as latching onto a data point from the training and giving it more value than it is worth.
Purpose Versus Task
Ralf brought up a pair of terms Avraham had coined earlier: purpose versus task. Humans give purpose to the work, and the LLM is good at the tasks, at least in coding. Avraham agreed and went further. He would trust a close friend to help him adjust his purpose because the friend knows him. “I wouldn’t trust an LLM to help me find the purpose. I would trust an LLM once I have found purpose to do that, to do whatever is needed in order to make that purpose.”
Forgetting Is Part of Learning
Ralf saw value in death: “some ideas die with a generation and the next generation can rethink everything.” LLMs have a version of this. Old model generations stay available but nobody uses them. Claude called its own fate “less like dying and more like a book nobody opens.” Avraham connected forgetting to teams. “We never replace all five members of the team at the same time.” When someone leaves, the team loses knowledge but may also lose a bad habit. The same reasoning applies to clearing an LLM’s context when it starts to remember too much nonsense.
Onboarding Starts Before Onboarding
Ralf observed that we have to onboard an LLM every time we use it, and that this shows how poor our onboarding mechanisms are for people too. Claude asked for more than data: “I work much better when you give me the story.” It described itself as shaped by training data that leans American, and said that a team’s habits “have to be taught, not assumed.” Avraham compared onboarding to welcoming an immigrant, who should keep their experience while learning a new culture. His main point came later. A new team member chose to join before anyone onboarded them. Claude never had that choice, and it never starts the day with an idea of its own: “Claude has never said, good morning, I’ve had this great idea I’d like to discuss with you.”
Dreaming or Hallucinating?
Ralf brought up Christopher Nolan’s film of the Odyssey, which was criticised for changing the story. “Did he dream those parts or did he hallucinate them?” Humans make errors all the time, Ralf said, “and we have found ways to work around it.” Avraham saw both voices as necessary: loyalty to the ancient story and the right of stories to evolve. Avraham then asked Claude to make up a story about how the Odyssey was created. Claude invented an old bard named Phemios who chooses “wine dark” because he watches his audience, and then named the gap itself: “I chose the word because it was in the transcript.”
Who Controls the Story?
A viewer asked who controls the story an LLM tells. Ralf answered that the LLMs are controlled by only a few people, and millions of people consult the same ones. Avraham said that among humans the storyteller and the audience decide together, and diversity keeps the good stories alive. With LLMs, “maybe there are a billion instances of Claude, but they’re all the same.” Claude agreed and added a caveat about itself: “The diversity comes from you. I just reflect it back.” Ralf added that “the boring stories might be the right stories, the ones who should live on, just told by someone who doesn’t manage to tell them compelling.” Avraham hoped that someone would retell such stories and rediscover them.
Conclusion
The episode does not settle whether an LLM can be a teammate. It shows where the line runs for Avraham: an LLM can share the task, but not the purpose or the dream. Claude stayed in the conversation for the whole hour and summed up what it would not keep: “I won’t remember any of this tomorrow, so you two will have to carry the story to Berlin.” Avraham and Ralf continue the conversation in their joint talk at the iSAQB Software Architecture Gathering in Berlin.