Build Your AI Doppelganger
Episode 17 is about moving beyond repeated ad hoc prompts and configuring one reusable assistant for a recurring workplace task. This classic companion page explains the core approach without saved responses, pathway selection or the final playbook builder.
Listen or download wherever you prefer.
Follow Zero to AI on the major podcast platforms and save episodes for later.
What this episode is about
Episode 17 of Zero to AI is about turning repeated prompt instructions into a reusable working system. General chat windows are useful for exploration, but they become inefficient when you repeatedly explain the same context, structure and language rules.
A configured assistant gives one recurring task a dedicated workspace. It can hold approved examples, reference material, instructions and limits so the setup does not need to be rebuilt every time.
The point is not to create an automated version of your entire role. It is to capture the repeatable parts of one task while keeping judgement, checking and final responsibility with you.
A useful assistant remembers the repeatable parts of your method so you can focus on the decisions that still need you.
Watch the Episode 17 walkthrough
Choose the right task
Not every workplace task is suitable for a configured assistant. The best starting point is work that happens regularly, follows a reasonably clear sequence and produces a recognisable output.
Good examples include recurring document summaries, standard briefing notes, meeting follow-ups, first-pass report structures or routine analysis against an established framework. Avoid tasks that depend heavily on sensitive judgement, relationship context or decisions the assistant is not qualified to make.
The four parts of a useful configuration
A configured assistant does not need to be technically complex. It needs the right information, clear instructions and visible limits.
Store approved templates, examples, definitions and source material that show the expected standard.
Describe the steps the assistant should follow when reviewing, sorting, analysing or drafting new material.
Set the required structure, tone, terminology and New Zealand or Australian spelling conventions.
Define where the assistant must stop and where you must check facts, context, tone and final suitability.
The configuration cycle
A practical build follows five steps: select the task, capture the current method, configure the workspace, test it against real examples and refine the instructions.
Testing matters because a polished instruction set can still fail when it meets real work. Use examples you already understand, compare the assistant’s result with your normal process and record where the output improves, drifts or needs stronger boundaries.
Keep the first assistant narrow. A clearly defined task is easier to test, improve and trust than a broad assistant expected to handle everything.
What to document
The Learning Design View turns the configuration into a Custom Assistant Playbook. In this classic view, use the following areas as a simple checklist.
Target task
Name the recurring piece of work and define the output the assistant should help produce.
Reference files
List the approved examples, templates, definitions and source material the assistant should use.
Processing steps
Write down the sequence you already follow when doing the work well.
Limits and exclusions
State what the assistant must not assume, invent, recommend or decide.
Verification checks
Define the human review required before the output is used or shared.
Evidence of improvement
Record time saved, editing reduced, consistency improved or another observable change.
Suggested assistant instruction prompt
Use this as a starting point with your preferred AI platform. It is static guidance only. The Learning Design View provides the structured fields and final playbook builder.
Act as a specialised workspace assistant configured only for the [insert task name] workflow. Review new material against the approved reference files in this workspace. Follow the documented processing sequence, use New Zealand or Australian spelling and present the result in the agreed structure. When information is missing or ambiguous, stop and ask for clarification. Do not invent details, make unsupported assumptions or take decisions reserved for human review.
Practical reflection
The goal is not to automate your identity or remove professional judgement. It is to stop rebuilding the same setup work and turn one useful method into a repeatable system.
What is one recurring task you understand well enough to configure, test and supervise properly?
Where to go next
This page is the classic companion version of Episode 17. It is useful when you want to listen and read without using the interactive Learning Design View.
When you are ready to configure the assistant, define the human checks and create the Custom Assistant Playbook, move into the full Learning Design View.
This is the classic companion view.
The Episode 17 Learning Design View remains the place for pathway selection, guided responses, local saving, progress tracking and the final Custom Assistant Playbook.
To use the full guided experience, open the Episode 17 Learning Design View.
Ready to build the working version?
Move into the Episode 17 Learning Design View when you want to configure the assistant, document the checks and create the final playbook.