Subscribe to Hard Mode First
>600 subscribers
Subscribe to Hard Mode First
>600 subscribers
Share Dialog
Share Dialog
Over the holiday break, I took three weeks away from my typical work duties, which included a full six days of digital detox (no phone, no Internet, no AI...a big deal for me!) Since I work fractionally, I didn't have any single person to pick up the pieces for me across all of my projects. This felt like a perfect use case for an "out of office" custom GPT helper bot.
I named her Nicole, and you can talk to her here. (Though admittedly her knowledge is a few weeks out of date.)
In this post, I'll share a bit more about how I built this OOO chatbot and what I learned along the way.
You can use any number of AI platforms to create a custom GPT or chatbot interface. I experimented with a few options, but ultimately decided to just go with ChatGPT, in part because I was already so familiar with the platform and also because I already had such a deep context window.
Here's what I included in the training prompt and instructions:
The Work Stuff.
First I started with the normal "out of office" reply stuff. I uploaded recent documentation from a few initiatives at work and hard-coded the names and emails of peers who could answer questions about each project in my absence. So, let's say you wanted to know more about Decoded Futures, the nonprofit initiative I helped start that prepares education and workforce nonprofits for the age of AI. You can just ask Nicole, and she'll point you in the right direction.
The Personal Stuff
Due to my prolific use of AI, my personal instance of ChatGPT knows a lot about me. So I started by asking ChatGPT to list out all of the things from its memory and build that into a list to train another AI to learn about me. (Meta, I know.) I also uploaded a recent copy of my resume and cover letter for some more context. Since I'd recently gone through an exercise of analyzing 100+ recent blog posts, I also used that as training data, both encouraging the AI bot to answer questions based on things I've already written about and also to point people toward blog posts that might be interesting to them.
Finally, I'm always in-between projects, I also wanted to make sure I wasn't missing an opportunity market and promote myself to a possible new audience of client base. So I fed Nicole a lot of details about my current projects, and also some of the ideas I was thinking about going early into 2025.
The Fun Stuff
I decided to have a little fun and tried to give my bot, "Nicole" an personality with a few Easter Eggs. But I won't spoil the fun for you. Go ahead, ask her about herself.
People don't click into OOO messages. I think most of my network simply saw the subject message from the automatic reply, or it went directly to spam, so they didn't engage. A better version would have been a non OOO triggered email responder that you could engage with via email rather than need to navigate to ChatGPT. I messed around with Lindy a bit earlier last year to try a version of that, but the time spent to set up temporary workflows didn't seem worth it.
There's (still) a general lack of familiarity with how to engage with a Chatbot. When I did invite a few of my close friends and family members to ask Nicole about me, I was often met with blank stares. We haven't really normalized the idea of how to quickly assess what a particular AI or chatbot can do for you (and what it can't).
There's a character limit in custom GPTs. Having made 50+ custom GPTs with OpenAI, I'd never before hit the character limit in the instruction panel. This proved to be an interesting constraint as I had to continually refine the instructions to trim down the details. With a bit more time, I would have made the character, "Nicole" a bit more fun.
Creating this "out of office" AI helper bot was an incredibly helpful, neatly scoped, exercise for me to learn more about how to define an AI-shaped problem and also get better at the "micro build." Just like publishing blog posts so frequently has vastly improved my writing game, conceptualizing and publishing so many mini-apps has helped me think a lot more creatively in other contexts as well.
If you're interested in learning more or building more with AI, I think the best thing you can do is to just start tinkering with all of the tools out there today and build some for yourself.
Over the holiday break, I took three weeks away from my typical work duties, which included a full six days of digital detox (no phone, no Internet, no AI...a big deal for me!) Since I work fractionally, I didn't have any single person to pick up the pieces for me across all of my projects. This felt like a perfect use case for an "out of office" custom GPT helper bot.
I named her Nicole, and you can talk to her here. (Though admittedly her knowledge is a few weeks out of date.)
In this post, I'll share a bit more about how I built this OOO chatbot and what I learned along the way.
You can use any number of AI platforms to create a custom GPT or chatbot interface. I experimented with a few options, but ultimately decided to just go with ChatGPT, in part because I was already so familiar with the platform and also because I already had such a deep context window.
Here's what I included in the training prompt and instructions:
The Work Stuff.
First I started with the normal "out of office" reply stuff. I uploaded recent documentation from a few initiatives at work and hard-coded the names and emails of peers who could answer questions about each project in my absence. So, let's say you wanted to know more about Decoded Futures, the nonprofit initiative I helped start that prepares education and workforce nonprofits for the age of AI. You can just ask Nicole, and she'll point you in the right direction.
The Personal Stuff
Due to my prolific use of AI, my personal instance of ChatGPT knows a lot about me. So I started by asking ChatGPT to list out all of the things from its memory and build that into a list to train another AI to learn about me. (Meta, I know.) I also uploaded a recent copy of my resume and cover letter for some more context. Since I'd recently gone through an exercise of analyzing 100+ recent blog posts, I also used that as training data, both encouraging the AI bot to answer questions based on things I've already written about and also to point people toward blog posts that might be interesting to them.
Finally, I'm always in-between projects, I also wanted to make sure I wasn't missing an opportunity market and promote myself to a possible new audience of client base. So I fed Nicole a lot of details about my current projects, and also some of the ideas I was thinking about going early into 2025.
The Fun Stuff
I decided to have a little fun and tried to give my bot, "Nicole" an personality with a few Easter Eggs. But I won't spoil the fun for you. Go ahead, ask her about herself.
People don't click into OOO messages. I think most of my network simply saw the subject message from the automatic reply, or it went directly to spam, so they didn't engage. A better version would have been a non OOO triggered email responder that you could engage with via email rather than need to navigate to ChatGPT. I messed around with Lindy a bit earlier last year to try a version of that, but the time spent to set up temporary workflows didn't seem worth it.
There's (still) a general lack of familiarity with how to engage with a Chatbot. When I did invite a few of my close friends and family members to ask Nicole about me, I was often met with blank stares. We haven't really normalized the idea of how to quickly assess what a particular AI or chatbot can do for you (and what it can't).
There's a character limit in custom GPTs. Having made 50+ custom GPTs with OpenAI, I'd never before hit the character limit in the instruction panel. This proved to be an interesting constraint as I had to continually refine the instructions to trim down the details. With a bit more time, I would have made the character, "Nicole" a bit more fun.
Creating this "out of office" AI helper bot was an incredibly helpful, neatly scoped, exercise for me to learn more about how to define an AI-shaped problem and also get better at the "micro build." Just like publishing blog posts so frequently has vastly improved my writing game, conceptualizing and publishing so many mini-apps has helped me think a lot more creatively in other contexts as well.
If you're interested in learning more or building more with AI, I think the best thing you can do is to just start tinkering with all of the tools out there today and build some for yourself.
No comments yet