A New Assignment: using ClothesPilot to pack for MVP Summit

I forget exactly who it was that came up with this idea but I think it was either Kevin or Zoe from the excellent Copilot Connection podcast: how about using ClothesPilot to choose my clothes for the MVP Summit?

If you haven’t been following, ClothesPilot is my Declarative Agent for Microsoft Copilot that uses APIs about my clothes, the weather and my schedule to pick what I should wear each day. I’ve been posting my results on LinkedIn.

MVP Summit is an annual meet up of Microsoft MVPs from around the world, hosted by Microsoft on their campus in Redmond. It’s a fantastic opportunity to meet and learn from some of the best in the industry, as well as find out from Microsoft their thoughts and plans, during a friendly, busy, exciting but NDA-covered week.

Going to MVP Summit always gives me a unique feeling: part excitement, part curiosity, part Imposter Syndrome. If I’m entirely honest then I tend to choose my clothing for this week focusing on comfort and blending in, not standing out.

So I was a little uncomfortable thinking about asking ClothesPilot to choose my clothes for me. But, I was really curious to see how good a job it would do, and how much modification I would need to make from what’s already there today.

Changing Purpose, not Code

I’ve said before that, for me, one of the big benefits of using a LLM is the ability to change requirements and do things that the original creator didn’t have in mind. I had absolutely no thought when I created ClothesPilot that I might use it to plan a week-long event, everything was centered around one set of clothing for the next day.

The two areas that I knew for sure would be a problem were the schedule and the weather. In part this was a problem that I created through my use of an “abstraction API” to make things easier. Because of this, I had API calls for “GetTomorrowsWeather” with the location set to where I live, and “GetTomorrowsSchedule”. Although the down-stream APIs which these called could be used to get different dates, they were set.

I decided not to worry about the schedule because the Summit sessions don’t go onto my calendar anyway. That just left the weather. I decided to create a single new API endpoint: GetWeatherSeattle10Days which returned a forecast of the weather in Seattle.

I updated my Declarative Agent manifest to tell it about this new API endpoint, but did not add any further instructions regarding the Summit. I wanted to see how versatile it could be.

Just to test that I’d hooked up the API correctly, I asked for a Seattle weather forecast. As expected, the weather. Not expected: snark about the Seattle weather! ☔

In Conversation with ClothesPilot

So, a Declarative Agent that has instructions around choosing clothes for one day, but with the ability to get a weather forecast for 10 days in Seattle. How would this fare?

Turns out, really well.

For a first-pass, this was quite impressive, given the lack of developer input / coding.

The thing I’m still learning about Copilot (and LLM chats in general) is that it’s not always a one-shot thing. It’s not the end of the world if the first answer to your question isn’t perfect, it doesn’t mean it’s a failure. Modify and improve. In my case I wanted a summary of the weather forecast for each day, and I there were clothes there that I didn’t think were appropriate (grey joggers and an oversized hoodie, more gym wear than corporate HQ wear!). But, we did a bit of to and fro, and landed on a schedule that I’m happy with. (which you will see next week as I post each choice on LinkedIn!)

Then, a game-changer…

This makes it so easy to pack my suitcase now, without worrying I’ve forgotten something. I love it!

Unexpected Benefit – No Overpacking

Whilst I was reviewing these choices I noticed that ClothesPilot was re-using some clothes on multiple days. This is not something that I’ve ever told it to do, and it’s been on my todo list to have a think about how best to address this for ClothesPilot’s normal behaviour, which is to mark clothing as worn, which then excludes it from selection for 2 weeks (to allow to washing etc). That’s a helpful way to stop it choosing the same combination of clothes every day, but doesn’t quite reflect the reality that I will sometimes want to re-use some clothes multiple days, especially if I’m working from home.

I was really intrigued by this. Why did it choose to do this? I’m not against the idea, and actually it’s very sensible for a week-long trip. My default when packing for trips is to overpack and then not wear a lot of the clothes I bring. Could this be a way to help me break this habit?

I wanted to understand more about the decision making here, but without changing it. I’ve found in the past that asking about why decisions happen with an LLM can lead it to interpret that as criticism and change behaviour, which I didn’t want. So, I gently tried to find out the reasoning in my most LLM-friendly voice:

That’s pretty hard to argue with. Left to my own packing I would definitely pack more clothes than this, but for this trip I’m going to go with the suggested outfits and we’ll see how it goes.

Conclusion: I’m pretty impressed

So, am I going to use the suggested outfits next week at the MVP Summit? Yes, I am!

If you’d like to see what they look like, I’ll be posting daily on LinkedIn, so follow me there.

If you’d like your own Declarative Agent (for clothing or something else) then check out my 5-part video and blog series on how to do this.