How I Built A Sales Co-Pilot App In ~20 Minutes
The rise of "code composers": How I used Replit + OpenAI/Anthropic/Perplexity APIs to help make me more productive in pre/post call GTM tasks, as well as call coaching
In sales, there are many repetitive tasks that you need to do throughout the day. A common flow below:
Pre-call: Before you get on a call with a prospective customer, you need to know what they do/how they make money/what they care about, as well as understand how this aligns with the problems your business solves. This involves desktop research, AI tools you need to prompt to make effective, and data synthesis to assemble the pieces.
During-call: You need to take notes and actively listen on a call (shout out to Granola, a game-changer tool to help alleviate the burden here).
Post-call: To keep the momentum going, you must also send a detailed follow-up via email or Slack. Doing this all manually is a headache.
Coaching: After a call, you always want to improve. However, getting a human to review each call and show you where you spiked and where you need to dig in deeper or do better is difficult. You could do this with ChatGPT/Claude, but you would have to go to another screen and re-prompt it for what you want.
Individually, these are easy but tedious, and the more calls you have, the more task switching makes you less efficient for many of these mundane tasks.
Because Granola helps me with the “during call” tasks, I’ll focus on numbers 1, 3, and 4:
To reduce task switching, I found it better to create my app or front end to input the things I was commonly prompting or needing to speed up time.
To do this in 2010, I would have needed developers to help me build this. While I’ve worked on becoming more technical over the years and more recently (I just finished CS50x), 2024 was an unreal year for AI app development.
To get from 0—>1 with an app, meaning to:
Build a sleek front-end
Implement logic, prompting, and call APIs from different LLMs based on the use case
Spin up a database to store all my information
This can be done in a matter of minutes. It’s insane.
I learned about all this from Greg Isenberg’s tinkering videos & Aman’s article with Lenny. After watching one, I was hooked and wanted to try it.
To build this app, I broke it down into the 4 core components I thought I’d need and matched them with the right tools:
Front-end: For the web app's UI. While I'm a Webflow enthusiast, I chose Vercel for its modern vibe and sleek interface options
Deployment: For getting the site live. Replit was my go-to since I'm familiar with its deployment pipeline
Development environment: I picked Cursor (so hot right now) as my IDE for its AI-powered features, which I figured pairs well with Vercel/Replit workflows (Replit has a native integration with Cursor)
LLM integration: I had the OpenAI/Anthropic/Perplexity API docs up next to me
I was tinkering with this tech stack for a couple of hours, and it was going fine until I found out about Replit Agent. I’m not the only one who feels this way; check PG’s tweet from last week:
What a freaking tool this is. It combines the front-end, deployment, and IDE use cases into one. This is plenty for my sales co-pilot use case, which is rather basic (yes, a classic “chatGPT wrapper”).
Note: I can see an SWE or developer looking down on this approach and only finding valor in “coding” if you are directly in the IDE and using a tool like Cursor or Codeium. I get it.
My stack to build this looked like the following (yes, re-using this image above, sue me):
One call out - why these APIs? Based on some customer interviews with developers and also my own experience, I have found that:
OpenAI: best for multi-hop reasoning and deep analysis
Have heard good things here from the Google/Gemini releases and will look to try this next
Anthropic: best for writing and coding
Perplexity: best search API to pull recent news off a Google search
Exa.ai I have heard from my network as an up-and-coming API that is just as good (but doesn’t have the brand name). Will be trying that next week
Okay, now what should I actually build? As mentioned above, I want a sales co-pilot tool to streamline my pre- and post-call tasks, along with a sales coach I can ask to help me improve.
A lot of what I took inspiration from is the work Dustin is doing over at ChatAE. Check the screenshot of his app; we’ll make something similar:
See here for a Loom video of the steps (first time doing this, please no “face for the radio, voice for the paper” jokes - I’m vulnerable right now, guys) of how I did this with Replit and the LLM APIs mentioned above. I figured doing this in a video would be better than writing it with screenshots.
I’ve been using this for a week or so, and it's incredible that we can now create apps like this. I can’t wait to see what's next and how fast this space continues to move.
Would love any feedback on how others are using AI and feedback to how I went about it all.
Thanks for sharing!