Designing half of a conversation is at a fork in the road, with two main options: building toward a real understanding of what a person is saying and generating the right response, or creating controllable, scripted responses to a fixed set of commands. The former approach is challenging due to its reliance on artificial intelligence and can lead to creepy sci-fi futures, while the latter offers more predictable but less nuanced interactions. A conversation model called CLEAT, which emphasizes context, shared language, engagement, agreement, and transaction or action, provides a framework for designing conversations that prioritize understanding the other side of the conversation. By focusing on these elements, designers can create software with conversation at its heart, fostering human-like relationships between teams, customers, and technology. The key to better conversations lies in understanding context, reaching agreements, and delivering the right answer in the right way, while also making sure the bot has a job to do and there's agreement about its role. Allowing for imperfection by narrowing constraints and enriching context signals can help mitigate mistakes and build trust between humans and technology.