I built a custom app and ChatGPT made it obsolete
Actually delighted about it because it opened my eyes to the future of AI-native software
I spent two weekends building a custom app to automate my team’s launch workflows. ChatGPT released a feature that made it irrelevant in 30 seconds. And I’m thrilled.
The problem to be solved
One of my goals with AI is to automate the menial tasks. A very reasonable place to start is highly repetitive and frequently occurring projects with a consistent, defined set of outputs.
Enter, our monthly feature launches.
Now these are what I consider to be “hygiene launches”. Like brushing your teeth, it’s good practice to constantly be communicating about product improvements. Every month there’s a defined set of assets that go out summarizing what we released to customers and the market at large — a blog, two sentence snippet in our customer newsletter, and a social post. Engineering produces documentation and release notes which marketing assets will link out to so not my responsibility to create.
First, a bit about the dope ass custom app I built.
Custom app #3: Launch Asset Generator
Custom app #1 here and custom app #2 here.
Tools:
Claude Code (my #1 love for vibe coding)
ChatGPT (the ultimate debugger) & ChatGPT API (content generation)
Supabase (for a lightweight database for CRUD operations on my custom prompts)
Vercel (for deployment, Vercel v0 for vibe coding doesn’t work for me)
Github (for hosting the code repo)
Beyond Identity NextAuth integration (for passwordless authentication, full disclosure I work here so dogfooding)
Linear GraphQL API (for pulling product / eng tickets)
What the app does functionally:
Uses Beyond Identity
Allows the user to select the right Linear ticket with feature description and value propositions
Allows the user to save, edit, delete, and use custom prompts for marketing blogs, emails, and social posts
Generates marketing assets with ChatGPT that uses information from the linear ticket and follows the custom prompt
Ta-da I’m never writing a feature launch email again!
What the app does technically:
Integrates with the Linear GraphQL API to pull information out of Linear and into the app
Uses Supabase to store and allow for editing, deleting, and updating custom prompts for marketing assets
Integrates with the ChatGPT API to combine Linear ticket and custom prompt for asset generation
Here’s what the app looks like (simple but cute):
Estimated time saved: 5-7 hours / month
Cost: $36 in Claude Code tokens
Then comes ChatGPT with a native Linear Connector
And it instantly made my application unnecessary. The functionality is the same, pull from Linear, user adds prompt, and asset gets generated.
Womp womp, right? Except that’s not how I felt at all.
I felt SO RELIEVED that I didn’t have to maintain my custom app and deal with all the feature requests.
There I said it.
Vibe coding is still deeply gratifying and fulfills my never ending need to tinker. But I find app maintenance and optimization tedious. Thank god I’m not an engineer huh?
Plus, how fucking cool is it that ChatGPT built the EXACT feature I wanted? And I’m hopeful that ChatGPT will continue to add to their integration features so I can automate even more workflows natively.
Seriously, use these ChatGPT Connectors. It unleashes the power of o4 and o3 to systems external to ChatGPT — I’m impressed and will write about the segmentation exercise I did with it. Here’s the instructions for the Linear connector.
My prediction: AI tools will allow users to remix their functionalities
By prediction I mean this is me manifesting my hopes and dreams.
You know how it’s standard practice for devs to fork someone else’s repo to build functionalities on top. Then, if it’s stable and useful, the fork gets merged into the main branch for all of consume.
Wouldn’t it be cool to remix your tool and tailor it exactly to your needs?
Today, you fork a repo. Tomorrow, you fork a product.
This is the mental model of AI-first products I’m betting on (and wishing for).
Today, SaaS tools scales by shipping a universal platform. But what if you could let every user tailor the experience to themselves.
For example: I build an AI-native CRM.
It has all the nuts and bolts for contacts, opportunities, campaigns, etc. I’ve also future-proofed it by making it API-first.
Now, instead of feature requests (and it’s evil twin, feature bloat), I integrate some sort of AI coding tool with a UI wrapper.
The admins has a button that says “Build your own feature”. On click, the admin gets dropped into “Remix mode”
In Remix mode, the admin is able to fork parts of the product for custom functionalities without knowing how to code or anything about my platform’s API endpoints.
Then if the admin chooses, they can submit the fork for review and merge into main branch. Or keep it proprietary, whatever floats your boat.
To some extent, this is already what you see OpenAI doing with Custom GPTs. The problem is Custom GPTs are super bare bones.
I want composable and customizable software and I want it today.