In this age and era of AI assisted development, there are tons of things you can build with it now that you perhaps didn't have time to work on before floating around in your head.
Like many who have picked up LLMs for assistance in doing building of projects, I have a graveyard of projects deep that haven't shipped for various reasons. As deep as my projects folder is with projects I have shipped, the list is long for the failures to ship.
Today I want to talk about those projects that haven't shipped. A sort of look behind the curtain so to speak. Each project that I've worked on has taught me things, sometimes more than others, every project was a lesson in something.
I guess you might call this the AI slop I never shipped?
Projects that I was so stoked for I bought a domain or two
SupportAddict
Chris Oliver and I worked on RelationKit.io for a few years (from like 2021-2025 end) together and while I largely built out the platform side, I wasn't able to give it enough time to make it go widely available. Part of the learning from RelationKit was that I was trying to do too much with it, so it wasn't good enough in any one area. Chris and his team for GoRails used it for support for quite a few years until recently replacing it with their own support tooling. So SupportAddict was going to be me revising the concept of a rock solid support tool for indie devs who couldn't stomach the Intercom costs.
I had a decent amount of the Rails platform build + some iOS application work done on it before I shelved it. I shelved it due to the sheer amount of work that would have been left.
PingTrail
One of the things that RelationKit did really well was allow applications to send API events and we'd wire it up into the customer profile, so think like "user created server" or something that a hosting company might want to track events wise. When I was pairing down RelationKit into separated projects this was one I was highly motivated to make.
It honestly was 95% done, and needed billing figured out before I shelved it. It was one of my first forays into letting LLM agents do more of the heavy lifting code wise when I would plan things and then let it rip, but the models were not nearly as good as they needed to be for that.
PopNote
PopNote was going to be a place where users could submit 'notes' on a popup on your site for feedback and issues. I had a hard time finishing this because it felt like it would have decimated the feedback portion SupportAddict might've handled... but also I was hoping to be able to target people who maybe only wanted the feedback portion and not the customer support side of things. This was shelved after a lot of the design iteration was done.
LogPop
This was either going to be the original name for PingTrail, or a project that would've ingested logs and 'popped' up the important ones to your team. Shelved just past the design phase.
UserWrite
I was tired of things like Medium for blogging and considered making my own blogging platform, paywalls freakin suck. Not much work on this past design iteration and MVP prototyping.
SpecGoblin
At a previous job we were using an AI startup to try to ingest our codebase + human written problems and break them down into a product requirements document. This was me experimenting to see if I could implement it. I have a Rails app that can successfully take a few sentences and break them down using LLMs into bite-sized specs/tasks that LLMs can iterate on. I shelved it because it felt like 1 of 10 pieces you'd need to fully iterate on the process.
FormGoblin
Form builders, like typeform and whatnot. Seems absurd to pay that pricing for when you need a few forms very rarely. App was built but it wasn't a good UX and needed more time to build out.
PhasePilot
I was super curious about using LLMs to help me learn a new language for reading/writing/speaking. A sorta new take on Duolingo of sorts that would have let me try having conversations with an LLM in the new language. Didn't get much past design iterations/mvps though because I realized how hard it would be to implement.
ByteUI
I was going to make my own design/component system using custom elements. I had some initial prototypes up, and was rebuilding them, but to do it right you need to know what you're doing for accessibility and whatnot. I kept finding myself making the same applications over and over and was getting frustrated with the copy-paste.
OneResume
This one was originally going to be a one-page resume portfolio site, I had wanted one that could keep my resume updated as I needed it and then generate PDFs for me I could use to send off to job applications and whatnot. I have a prototype of the app built but never hit a production deployment.
PlayNeedle
I was working at a company that did music licensing and really into the idea of building my own Apple Music / Spotify music client. I have an entire Svelte, Electron application for this, but stopped short of the backend that would let you auth with Apple Music. It would have targeted Linux and Mac at launch with Windows to come.
Things that didn't make it off my computer
FetchKit
I was going to rebuild Postman, because it felt like something I needed while I was working on rewriting an API for a previous job.
A Mac App Launcher
This was a more ambitious project of mine, I wanted to replace Raycast and Alfred and Spotlight with a custom built Mac launcher.
My own Terminal Emulator in Rust
I was going to try to rebuild my own terminal emulator in Rust. Shit is hard to do right.
App deployment tool
What if Kamal had a UI? Basically was trying to replicate all the stuff I love about Hatchbox. I shelved it after design iteration.
AI Management Platform
My first attempt at what would later become Atlas Workspaces, but not anywhere close. Would basically stream in logs from Claude sessions to a web UI.
BetaQueue
Email lists but for app beta signups. I was working on like 15 different ideas and was trying to think of ways to manage signups. I invented work for myself that wasn't a real problem for others.
Centinel
I was going to make a personal finance budgeting app. Some work was done towards this but I stopped short of getting it connected to my transition history.
Changelog Recaps App
I was working on an app to take all of my projects history and then generate changelogs from the commit / PRs. Using AI to summarize it and the human to approve it before posting.
CI Actions Platform
Initial prototypes of an Actions runner like GitHub actions and whatnot. I believe actions had had some issues and I was frustrated and rage coding.
Codebase Analysis
I was using RAG and PGVector to break down codebases to feed it to LLMs to ask it questions about how the Rails applications worked. Neat concept but I didn't push it too far.
DNS Tools
When you have 80+ domains, you need a way to manage them all and see when they're renewing. So I was building scripts/a UI to see and manage them.
Email Client
I was like I have 10+ emails, what if I had one inbox for them all in my UI. Just did the design work on it before shelving
Email Editor
What if there was a clean way to design emails for all my dad gum projects? So I could use a WYSIWYG editor for them and drop in the components. It had a ViewComponent piece of it that I could compose together to get the UI that'd be consistent for every email client across the board for tables/css.
Email Service Provider
I was really on an Email kick, huh? Think a pretty UI onto of Amazon SES or another provider that sent the emails (since that's a ball of yarn I didn't want any part of post my experience seeing the hassles at a previous job). It had design work and MVP planning for the application architecture but not real progress made.
Error Monitoring App
Turns out it's really easy to want to build your own tools, but hard to execute them. Honeybadger does like 90% of what I want, AppSignal does like 120% of what I want, this would have done exactly 100% of what I wanted, but was hard to follow through on.
Incident Simulation Game
I had a fun time building this one, it would have launched as an app that would generate a like Incident simulator for production incidents, perhaps showing database issues that a team would need to coordinate and debug via it, then recap how they handled incidents post-mortem and whatnot. Help break an ice and get the conversation going about how to handle incidents on production-like systems on engineering teams
Nation Simulator Game
Before I got into coding, when I was still on satellite internet, I found a Nation States Game that was a solid 2-3 year period of my high school life. I was curious about recreating that feeling, but it didn't land the same. I guess that's nostalgia for ya. Sometimes it hits, other times it misses.
Reddit Client
I'm still pissed about Reddit killing Apollo and was going to build my own client for it. Design work was good but then I needed to build it.
SpellPing
Haha, I forgot this one existed. I was building something that would run 'spells' on API endpoints and ensure they 'pinged' in the proper expected time. Think endpoints that returned within SLAs and whatnot. Did they include the expected response JSON ids and so forth. This was spurred out of a production incident at a previous job where our API responses stopped sending back expected data.
Status Page App
I was going to build a status page / uptime checker app. Design work was done but none critical work was made on it.
The graveyard taught me a lot
I learned a lot from each of these failed endeavors. Most never made it to a production deployment, heck most never made it into a git repo. The ones that did make it to a production deployment but haven't gone anywhere since are a story for another day.
Experimentation and Prototyping still can teach you a lot in this day and age of AI.
I see a lot of valid criticism against LLMs that people are saying it hasn't helped large companies ship better software faster, and I think that's fair. But it can teach you a lot about mental model shifting and how to work with it better when you do experimentation and prototyping like I have done.
My day job work with LLMs has gone significantly better because of my prototyping and experimentation outside of work than had I not done it. I know how to ask for things that get better results. This leads me to shipping better software faster at work because of it.
Anyway, don't be afraid to prototype and experiment, chase your excitement when you can. Embrace the chaos inside of you.
Thanks for reading everyone ❤️