If only I could clone myself!

Note: This is a summary of the presentation I gave at the Australian Institute for Training and Development conference in May 2025.

Especially for those solo L&D practitioners in an organisation, the idea of having an few extra pairs of hands is alluring! The good news is that some of the AI tools on the market now allow us to do almost that.

It is not about handing off your tasks to a generic LLM tool (such as Claude, Co-Pilot or ChatGPT), and asking it to do your work for you, although I’ve seen countless examples of people doing this, and treating LLM tools as if they were all-knowing entities that could magically spit out perfect work.

Instead, what we’ve started doing internally at Pukeko Learning Solutions is design custom GPT assistants that are able to perform very specific written tasks, with our very specific guidance.

Here's an example

If we’re tasked with writing some appropriate learning outcomes (LOs) for a project, rather than start with a blank word document, we can use our Learning Outcomes Writer GPT Assistant, which we’ve designed, to produce a first draft.

We have programmed the Assistant to:

·       determine an appropriate number of LOs for the length of course

·       include cognitive, affective and skills-based LOs,

·       begin LOs with a verb,

·       write LOs that involve higher-order thinking skills, as per Bloom’s taxonomy,

·       tailor the LOs to the content and the target learner group; their context, current skills, knowledge and attitudes.

How will the Assistant know those things at the bottom of the list there? It has to ask us. It has to ask the user a series of questions, until it has enough information to provide a first draft output.

Once we have a first draft of LOs, the user (that’s us!) can choose to take the draft and continue to work on it offline, or continue to interact with the Assistant, in order to refine that list.

At the end of the day, it’s the human user who is responsible for the quality, so there’s always a bit of back and forth until we’ve got the set of learning outcomes that we’re happy with.

What else?

That’s one small task, and a task that only comes up at the start of every project.

The trick is to think about how many other very specific tasks can you create an Assistant for? My current list has about 30 custom GPT Assistants in it. And each time I use one, I think about ways I can improve the programming, so that it is a smoother process or a cleaner output next time I want to use it.

It’s no different really to using a static template when you go to produce a document that you’ve produced many times before. The content might be different, but the thinking behind it, and the sections in the output are very similar.

Example of a template we've created and use in-house

We have many templates that we’ve created for our use inside Pukeko Learning Solutions, and now we’re working towards creating Assistants that can act like interactive templates for us, and produce first drafts much more quickly than we could ourselves.

Want to learn more?

To view the slides from the talk, which include short video demos for how to create custom GPT Assistants using the paid version of ChatGPT, click here.