From skeptic to believer - utilizing Chat GPT as a code buddy

Being the startup-ophile I am, I've known about / used GPT3 since it came out, and when Chat GPT was released, instantly started playing around with it, then purchased Chat GPT premium to access GPT4 a few months ago.

And while I pay for Chat GPT, for the longest time, I had an aversion to integrating AI tools into my workflow. It felt... impure, almost like I was cheating or not working "properly" if I was augmenting my ability. So although I was building things on top of chat GPT, like Docent or AskSEC, I never actually used it in my day-to-day.

Tools like Copilot were nice (but largely ineffective especially for larger codebases or complex logic), and various wrappers of GPT3 felt gimmicky / not super functional, so I largely ignored using Chat GPT on day-to-day for programming work.

That all changed a few weeks ago.

For the past few weeks, I've been programming a lot on top of Hubspot, Salesforce, and Slack APIs. If you haven't worked with these API's I don't recommend doing so. They're dense, the guides/documentation are sparse and often confusing (looking at you Slack API docs), and generally a complete pain in the ass.

The original idea for Docent was actually creating a Q&A wrapper around documentation, so I started interacting with chat GPT to write pieces of API code or ask about how parts of an API worked, and it worked surprisingly well.

While occasionally out of date, Chat GPT has been fantastic for actually exploring how an API functions, writing snippets of logic, generating types from docs or an example JSON, or chaining together API queries for some end goal.

And thus Chat GPT has started working its way into my everyday developer life. While not super useful for generic frontend engineering tasks, when tasked with writing pretty complex logic for a Slack Bot, it does surprisingly well (I've only had to correct it 2-3 times the last week) and has saved me easily 3-4 hours of work a week.

That being said, the experience isn't great. The native chat interface is clunky, slow, lacks shortcuts, and doesn't provide a great experience. And while there are tons of wrappers out there, I'm still on the hunt for the "Superhuman for Chat GPT" and have been surprisingly disappointed.

While I don't think GPT will steal software engineer jobs yet (it struggles with complex logic or large codebases), it has been very effective for saving me from writing monotonous code that adheres to a spec, and generally saving my sanity bits and pieces at a time.

PS - if you are building / using a great Chat GPT wrapper I will be in your debt forever ❤️