How skills, agents, and other AI features work

Agents! Thinking! Skills! Subagents! Cowork! AI products have become more sophisticated. Some might say more intelligent. We have a come a lot of ways from prompting to a chat-bot. Or have we? All of these ‘features’ are marketing speak for ‘prompting a Large Language Model‘, combined with conventional software. Sometimes not even that. Segmenting all of these things into different functions is quite useful for technical users. At the cost of alienating average users. In this post, I will try to demystify these features. By explaining how they are just prompts. AI Chat is a prompt… …that is very very long. Back in the days of GPT2(then the mOsT dAnGeRoUs model ever!!![1]), OpenAI launched a text completion tool. You gave it a prompt and it continued the text. Like autocorrect on your phone, but using a better technology. So if a model can extend any text, it can predict the next part of a conversation. You can learn more about how that works(in simple terms), in a previous post I…