Much of the recent work on personal agents assumes that the best agent is one you can hand a task to and leave alone. What this tends to overlook is context. An agent is only as useful as its understanding of the person it works for: what they are working toward, how they tend to go about things, and what they would want done next. Most agents begin each task knowing very little of this.
Many everyday tasks are also not well suited to being handed off entirely. People generally want to see what is being done on their behalf, many tasks need their input at some point along the way, and some can only be done on their own computer.
Otto is an attempt at an agent that works as an extension of you rather than as a separate assistant. It runs on your Mac and does its work on your screen, where each step can be seen as it happens. It offers what it expects you will want to do next, and if you accept, it carries the task out in front of you, waiting wherever a step requires your input.
Because it works alongside you, Otto is also in a position to learn from you. It gradually builds up a sense of what you are working toward, the tools you rely on, and how you prefer things to be done, and it draws on this each time it suggests what to do next.
The aim is that, over time, it comes to work much as you would yourself.