AI PATH PRO / TEXT EDITION
Read it and put it to work.
AI Path Pro · 20 minutes
Choose a recurring preparation task. Separate the stable method from the data that changes with every case. Use this to define a GPT, skill or plugin supported by your working environment.
The method
Choose a recognisable job
State when to invoke the assistant and when not to. Preparing a quote brief is a useful scope; taking over every sales activity is too broad for a first version.
Separate method from case data
Shared templates, checklists and good examples belong in the knowledge pack. Current requirements and dates are separate inputs. Give every maintained source a version and an owner.
Try three kinds of input
Use a complete case, an incomplete case and a contradictory case. Record the output you expect. Identifying a gap and asking a focused question can be the correct result.
The quote-brief assistant
Illustrative example. A sales coordinator extracts the same six fields from each enquiry. This case includes the requirement and quantity, but no delivery location or requested date.
Standing instruction: prepare a brief from the supplied case data and mark missing fields. Output: requirement, quantity, location, requested date, open question and source passage. In this case, location and requested date are “missing”. Next action: prepare two focused questions. The assistant does not invent a price or a committed delivery date.
Your worksheet
- Name and trigger
What single task is this for? Write a concrete starting request and one situation where it should not be used.
- Standing method
In what order should it work, which sources may it use, and how should it handle gaps?
- Knowledge and case data
What belongs in the shared knowledge pack, who updates it, and what must be provided separately for each case?
- Output and trial
Which fields are required? What should happen with complete, incomplete and contradictory inputs?
Review checks
- The knowledge pack and current case data are kept separate.
- Sources have a version and a responsible owner.
- Each of the three trial cases has an expected result defined in advance.
Save a working example
Keep the instruction, a good input and its checked output together. Give the pack a version and record what changed since the previous trial.