Two months ago, Katie Seiffert, the COO of Catalyst Opportunity Funds, wanted to know how to better use AI at work.

She had previously used ChatGPT and Claude for text generation — things like drafting emails, and refining writing. She had tried “one-shot” prompts that others had reported success with, but wasn’t in the habit of using LLMs.

In a short strategy call, we reviewed a few use cases together, and brainstormed how she could intentionally tackle these inefficiencies through appropriately using process redesign, automation, and AI. 

We showed Katie examples of what was possible in an Operations and Finance context, including a few “vibe coded” scripts, and sent her on her way.

Two months later, we got a status update from Katie recently on how she’s applying our AI strategy (spoiler alert: it’s going really well!).

Linda Mutricy (Crafty Crow)

What are you doing differently now as compared to two months ago, with AI?

Katie Seiffert

I've always thought about how to make processes better. Now it feels like I have this sparring partner with ChatGPT and Claude. 

I'll just describe the result that I'm thinking of. And because I'm non-technical, sometimes I have no idea if this is possible.

Before, I would have just written off my ideas. And now I'll ask ChatGPT, “Is this possible?” Or, “Give me 3 different ways I could do this.” 

ChatGPT would answer me along the lines of, “These are the different options for how you could connect these apps in your work.” I know my own processes better than ChatGPT does, but it gives me enough confidence to put in the effort to make my ideas a reality.

I've put “vibe coding” into practice at work, with a very simple use case of applying email labels. 

Up until this point, I could label my entire inbox, but I didn’t know how you could use those emails outside of Gmail. I wanted to create a database for my emails that I could easily refer back to in the future.

ChatGPT gave me options, like using Google Sheets or Airtable. It gave me reasons why either platform would work. I ended up setting up an Airtable database that populates when I label certain emails.

And then I prompted it, “Tell me exactly what steps I need to do.” 

I had to retrain myself to say, “Walk me through the steps”, or otherwise ask for clarification, knowing that it’ll tell me how to do it. It’s okay to not understand the setup instructions right away. You’re just talking to a computer program. 

Linda Mutricy (Crafty Crow)

You say it so casually, but since we've last chatted, you’ve created a working piece of code where, every time you label something in Gmail, it takes that email content and puts it into an Airtable database, so that you're able to then go through it at scale later.

Katie Seiffert

Yes, I used Google Apps Script, which you recommended in our previous catch-up. 

Linda Mutricy (Crafty Crow)

Can you walk me through your thought process when you wrote that prompt, and then you realized you had to implement that Google Apps Script code it had written? Was it intimidating?

Katie Seiffert

It was intimidating, initially. It told me how I could do this project, and I had that hang-up in my mind, where my immediate reaction was, “I don't know how. I need to learn Google Apps Script.” 

But then when I just said, “I don't know how to do that,” it was shocking how specific ChatGPT's instructions then became. 

Even when I was working on the Airtable integration, it told me that I needed an API key, but Airtable has now sunset API keys, and they've switched over to the Personal Access Token.

I thought, “Oh, this is a roadblock.” Then I just told ChatGPT, “Airtable doesn't have API keys. They’re using the Personal Access Token.”

It responded, “Oh, yes, of course. Here's what you need to do now.”

It was just shocking how much I didn't have to know, navigating this vibe coding process.

It definitely felt overwhelming, but I learned a lot. 

I had to change things in the script, because ChatGPT didn't initially create the best user experience.

The script originally removed the label after syncing the email. I still wanted to keep the initial label and just add a “synced” label.

So we had to iterate on that. Going through that iteration helped me learn about the script and what was happening, because I could see the clear directions and I was involved in the changes.

Learning is so much better when it's hands-on. Even though sometimes I was literally copying and pasting between two programs, being in the mix, I'm already learning what the underlying logic and code is doing.

Linda Mutricy (Crafty Crow)

What’s next for you?

Katie Seiffert

This experiment with Google Apps Script and the discussion with you has opened up massive amounts of possibility.

My job is not that complex. I’m only using a certain number of tools. It’s a lot of email, spreadsheets, and files.

One of the things I’m going to tackle next is internal requests. Right now, all of that is done within email. But drafting the email is cumbersome, requests are time-sensitive, and we want to make sure it doesn’t get buried in the approver’s inbox. I also want to introduce compliance in the process, tracking what was previously approved.

And so I'm thinking of setting up a Google Form that would capture requests. It’d populate that into Google Sheets and send an email notification for an approval. We also have a database of information that would go with every request, and that email notification would marry that information with the request. 

Being able to pull that information immediately with the request that’s being made, and then capturing the approval of that process, feels totally within reach to me. 

Even two months ago, I would have had no idea how to do that, or would have looked for one tool that could do it all.

And so that's where the real sense of possibility is coming from. I don't have to necessarily fit my work into one predefined application that exists out there.

I have the ability to spin up workflows for exactly what I need, that meet my team where they're already working, in the interfaces that they're comfortable with.

I could probably put together this whole process myself. And maybe the approvers are still only using email, because that's what they're comfortable with.

But behind the scenes, I can have all this complexity and make sure I capture this information compliantly. It wouldn’t be like having to manage change within a team, which is sometimes the hardest part. 

Linda Mutricy (Crafty Crow)

How do you think differently about using AI, now that you’ve accomplished all of this, vs where you were two months ago?

Katie Seiffert

I actually was just telling someone on my team today exactly your feedback, which was really direct, but also exactly what I think people need to hear.

Everyone is trying to make this massive leap from not having good, defined processes, thinking that AI is going to be our savior. 

Some of the problems that I expressed in our initial call were purely process issues that didn't need AI in particular. They just needed to be more thought-out and automated.

But then, what really added value was when you said, “You can use AI to help with that automation.”

So maybe AI isn't the tool that ends up being part of the workflow, but AI can be that partner for you to start making progress on defining those workflows. 

If you want to use AI chatbots for more than drafting emails — maybe even as a creative partner for workflow design, or a software engineering assistant to prototype those flows, we’d love to help. Book some time via our calendar link.