Outputs

What comes out of the solution (results, text, files)

Outputs are what comes out of the solution and into someone's hands: the result, text, file, decision, or action that makes the whole thing worth running.

Defining outputs before you build is one of the most practical moves in this cluster. When you know what you want to receive, you have a way to evaluate what the solution actually produces. Without that, review becomes guesswork: you are looking at AI output without knowing what good looks like.

Outputs should be described in terms a real user can evaluate, not in terms of what the AI generates internally. "A structured summary of the call" is an output. "A 200-token response" is an implementation detail. The person reviewing the output needs a human-readable standard to judge against.

It is worth being explicit about format too: where the output lands, what file or structure it takes, and who receives it. An output that cannot be read by the next step in the process is not an output yet.

Make it visibleWrite a one-paragraph description of an ideal output from the solution, as if you were handing it to the person who needs it. Include what it contains, roughly how long or large it is, and where it should land. Use this as your review checklist when the first real outputs arrive.

Why AI needs this

Each part of your work matters to AI in a specific way. Some of it is context a tool needs before it can help, some of it is work a tool can take on, and some of it is judgment that should stay with you.

Quality standard

You cannot review AI output reliably without knowing what you were expecting. A described output gives reviewers a benchmark before they read a single word the tool produced.

Handover to the next step

Outputs often need to go somewhere: into another system, another person's workflow, or a shared file. Defining the format makes that handover possible without manual reformatting.

Accountability point

When the output is clear, it is also clear who reviews it and signs off before it is used. Without a defined output, that accountability step tends to disappear.

Questions to explore

Use these on your own or in a group. There are no right answers, only better conversations.

What does the person receiving the output actually need it to contain, and in what form?
Where does the output go after the solution produces it, and does anything need to happen to it before it is used?
How will a reviewer know if the output is good enough to pass on?
Are there cases where the output needs to look different for different users or situations?
What should happen when the output is wrong or incomplete: who catches it and what do they do?

Readiness traps

An output defined as "the AI's answer" is not defined at all: you need to specify what the answer should contain and how it should be structured.
Outputs that go directly from AI into a downstream system without a human review step will eventually cause a mistake at scale, when nobody is watching.
If the output format is incompatible with how it will be used (wrong structure, wrong location, wrong owner), users will work around the solution instead of using it.