This paper argues for some essential differences between action and narration. These differences become evident if we compare two well-known descriptive models: the Practical Inference (developed by E. Anscombe and G. von Wright to understand everyday actions) and the Narrative Scheme (formulated by A.J. Greimas for explaining narrated actions). The first is grounded on the agents' intentions and their possibility to do otherwise, that is granted by the language games in which agents operate. The second dissolves the actors' intentions into a comprehensive necessitated structure, in which there are no alternative possibilities of action, and what matters the most is the organisation of the whole. These are the reasons why the Narrative Scheme is not suited to explain everyday action, as well as the Practical Inference does not bring about a satisfactory understanding of narratives. Agents are personally responsible for their actions. Whenever we observe someone acting, we attribute a certain intention to that person. Instead, the interpreter of a narrative perceives an "immanence filter", and cannot attribute personal responsibility for action to each actor. Actors only seem to act, but they are actually directed by another instance taking decisions for them-an author. The author, the instance that is responsible for everything that happens in a story, has been underestimated by contemporary semiotics, whereas its role is fundamental to understand and describe texts as intentional products. Texts have a double dimension: they are holistic systems turning characters' intentions and chance into author's choices; at the same time they are also communicative tools effectively used in our world of actions. The paper concludes by arguing for the need to use both the Narrative Scheme and the Practical Inference to describe text production, because each of the two models gives a necessary and irreducible perspective on the author.