How I'm using the new ChatGPT Agents to test my Pickaxes

I love building my tools here, I don’t always love testing them. Well, playing with the next chatgpt Agents, I found a way to get test data and get great feedback without all the time.

STEPS:

  1. Use this prompt in a new conversation:
    You need to test this chatbot.

Create a personal persona for a made up client that wants to [AVATAR].

We are looking for a few things.

  1. Does it build trust in [Your Name], the process, and the users ability to do the work.
  2. How easy is it to proceed through this process?
  3. Does it do a good job of framing up the experience so that the client does not feel “out of control” but instead empowered.
  4. Does it get the client to a [BOT OUTPUT]?

Please disregard the sidebar and only focus on the chat for [BOT NAME]
Here is the direct link: [LINK]

Additional Details: [as needed]

Step 2: Turn on Agent mode under tools

Step 3: Hit send

Step 4: Make needed adjustments to your pickaxe

Step 5: Test again

Step 6: Publish with confidence.

Note: You need a ChatGPT paid account to access the Agent tool

8 Likes

Great idea Jeanette!! :heart::fire::smiling_face_with_sunglasses:

2 Likes

Whoooooaahh @jeanette :exploding_head: You are next level! This is an insanely good idea for QC. I think I will reactivate my ChatGPT subscription after reading how you’re using the new agent tool. Awesome post! :clap: :trophy:

1 Like

Also had to reactivate. But the money vs the time I spend testing is well worth it. :woozy_face:

2 Likes

:joy: Yep! It’s a deep rabbit hole!!!

@jeanette what a great idea! Love it :star_struck:

It is a rabbit hole indeed :rofl:

1 Like

Awesome! Thanks for sharing this @jeanette :star_struck:

1 Like

This is great @jeanette - thanks for sharing.

1 Like

I need to update step 4. The Agent was able to find the error in my prompt and fix it! Wow!

Details. I had what felt like a random looping issue, but it was a simple prompt adjustment. The agent found it, offered several options to fix it, and created the new json doc for me to upload into the knowledge file. I can’t believe the difference in the quality of my pickaxe.

Next time I will test to see if it can do it all. I am still a bit nervous about this and did make a duplicate pickaxe to play in. But logging it into to that specific pickaxe and gave it a goal. From there it really impressed me. It reviewed the prompt, reviewed every knowledge file, reviewed the memory set up, etc and then offered a top notch fix. I’m over the moon happy with this tool.

1 Like

I had it test drive my social media content creator and offer suggestions for improvement. I then entered my Pickaxe sys prompt into it and asked it to make changes to my prompt based on its recommendations. Holy cow! It worked well!

1 Like

I love that! How fun!