What I learned

Building personal AI copilots is easier than I thought. Heuristically, treating AI systems as human systems gives a good idea of how to provide them with information and context. There are 4 parts to building an AI copilot (inspired by this post):

  1. Hire your copilot
  2. Onboard your copilot
  3. Kickoff an initiative
  4. Put your copilot to work

All the prompts that follow stem from the same post. I’m keeping them here for my own reference and will be adapting them to my own use cases. Which I will share in a future post.

Context

Copilots are created using the Projects section of your AI of choice. Allowing you to define specific system level information, knowledge and context that persists across chats.

Hire your copilot

Hiring is about understanding the role, skills, personality and behaviour you want your copilot to have. Like with humans, you want to think about the things that are best for the role not necessarily just for the task.

An example prompt:

I am a [role] at [company name], and you are my expert coach and advisor, assisting and proactively coaching me in my role to reach my maximum potential.

I will provide you with detailed information about our company, such as our strategy, target customer, market insights, products, internal stakeholders and team dynamics, past performance reviews, and retrospective results.

In each conversation, I will provide you with information about a particular initiative so you can help me navigate it.

I expect you to: ask me questions when warranted to gain more context, fill in important missing information, and challenge my assumptions. Ask me questions that will let you most effectively coach and assist me in my role.

Encourage me to: [list the values and behaviors that make you successful in your role]

I want you to find the balance of: [traits you want in a thought partner and coach that will be both effective and fun to work with]

Onboard your copilot

Onboarding is about giving your copilot the context it needs. In a work context, this could be strategy documents, product information, technology stacks. Anything that you’d expect a new hire to know before they start a task. These files can be uploaded to your project.

If documentation doesn’t exist, you can ask your copilot to interview you to help create it. An example prompt:

Please review what I've shared with you and ask me questions to help complete your knowledge.

What important information are you missing (company- and product-level) that you need to help me across all my initiatives, present and future?

Focus more on company, org, and industry level. Focus less on short-term conditions or initiatives (e.g. resources, constraints, or individuals), as those will change over time.

I want you to ask me these questions sequentially (most crucial questions first), so I can answer one at a time.

Which can then be followed by this prompt to help create the document:

Please create a document with the new information you've learned in our conversation. Only include new information learned in this conversation (i.e. that wasn't already in project knowledge). Don't outline any outstanding gaps in the document.

Optimize the document for adding to the project knowledge, so it can apply in all our future conversations.

Use exact quotations of my original words for the most salient and important things

Kickoff an initiative

When you start something new, try to use a single chat per project, keeping all context in one place. Then, use the voice-to-text funcionality to “kick off” the project by providing all the context someone might need on this project. I personally like using the voice-to-text funcionality in ChatGPT because I can ramble (I’m a verbal processor) and it can create all the structure it needs without me needing to provide it.

A starting prompt could be:

Now that you have the context on my company and my team, I want to tell you a bit about the initiative that I am working on and give you the specific initiative context.

This is the starting point of what I know, and I’ll be updating over time as more information and insights come in:

[Share what you know about this initiative at the start in any order.]

Put your copilot to work

You can kick this off simply by asking:

What is the single most important thing I should do next?

and see what it comes up with. This LLM now has access to all the context you’ve provided without the need for a long prompt.

Off the top of my head, some really cool use-cases could be:

  • Managing your team’s personal development.
  • Staying on top of projects in terms of meeting notes. Also to identify missing gaps that you have in your understanding.

A note on chat size limits

You may end up hitting chat size limits depending on which LLMs you use. To get around this, you can prompt the LLM to summarise the conversation so far.

This conversation has reached its context limit. Create a document that can serve as the initial context for a fresh blank LLM thread. Your goal is to preserve approximately 90% of the conversation’s value and context while reducing its length by ~90%.

Act as an expert handing off to another expert who will help me, and set them up for maximum success, as close as possible to having been there all along. Tell the new expert what instructions or behaviors to exhibit that I’ve implicitly or explicitly requested of you. Tell a chronological story.

Use your best judgment when it would be particularly valuable to use the original exact words of either the user or AI.

Skip any context that is already found in the project knowledge or in your system instructions, since this will be available automatically to the new thread.

Create a second, separate document with what you chose not to include, and why.

Key Takeaways

  • Projects are a great way to keep context localised for tools like ChatGPT.
  • Babbling to AI is a more effective way of giving information (especially if you are a verbal processor).

Source: https://www.lennysnewsletter.com/p/build-your-personal-ai-copilot