Episode 9 takes on the question the show was built for: whose words are they when the ghost is a machine? The guest is Dylan Latour, a ghostwriter who launched an AI-native agency in 2024 serving the mortgage industry, and who is blunt that writing is less than half the job.

The real work, he says, is pulling ideas out of a client's head and getting them brave enough to publish. Then the turn. Everyone blames AI for the flood of soulless content, and Dylan flips it: this is human slop, so stop blaming the AI and start blaming yourself. From there the three of them follow the thread through the creator economy, why the studios should fear AI more than the actors do, why an online presence is now mandatory, and a genuinely optimistic vision of a future where the boring work is automated and people do what gives them life.

What if AI is not a machine we built but an intelligence we are domesticating, the way early humans domesticated dogs 15,000 years ago? Chen Gu, an engineer who became a lawyer and now builds AI tools for the legal profession, brings a theory that reframes the whole debate. When humans tamed dogs, he argues, we may have domesticated ourselves in the process, and the same thing could be happening now with AI.

Andrew takes the optimist seat and sees a path to symbiosis. Liz presses on control, morality, and who gets to set the rules. Chen lands somewhere sharper: you cannot guarantee a moral AI any more than you can guarantee a moral human, and the only real safeguard may be individual ownership of your own model. A conversation about power, trust, and whether we end up as partners or pets.

Episode 7 takes on the topic Andrew has been itching to argue. The death of software as we know it. The guest is Eilon Shalev, CEO and co-founder of Elphi, the show's first guest from MIT, and a self-described business graduate who is not a software engineer.

That last part is the whole point. Eilon now builds end-to-end features inside his company's actual codebase, tests them himself, and hands a working product to his senior engineers to ship. Not a demo. Not a wireframe. Real code. From there the three of them follow the thread. If a non-engineer can build the feature, what is the engineer for? Eilon's answer reframes the job around architecture, judgment, and prompts, not typing code. Then it gets weird and fun: a hundred years out, custom LLMs, neural links, two societies, and whether you are the product.