Outcome, process, task, goal

Almost every AI project that fails does so by confusing these four. They are not synonyms and they are not interchangeable.

Outcome is what the company wants to happen in the world. "The sponsor renews the child sponsorship because they feel they are following the life of a real child." It is not verifiable, and that is exactly why it needs to be written down: it arbitrates the decisions the numbers cannot resolve.

Process is the path to get there. The annual correspondence. It has a beginning and an end, but it is too large to delegate as a whole.

Task is the delegable unit. "Write Ana’s letter based on this year’s report." It has an input, a cost, an owner, and a point at which it is done.

Goal is the verifiable instruction within the task. "180 to 320 words, in the child’s voice, every claim sourced to the report."

The entire chain needs to exist. Skipping the task is the most common mistake: the agent is asked to handle the process ("automate the correspondence"), has no completion criterion, and spins its wheels. Skipping the outcome is the most expensive mistake: the system becomes excellent at hitting targets that do not matter.