Not going to even get into the hype crap: Here’s the working code: https://github.com/startupsandcode/gmail-organizer
You will need to create a Google Cloud Project and OAuth Client (with) (not giving out my creds on that one, sorry). But other than that, this will work out the gate.
I have the full workflow and screenshots below, but wanted to talk a little bit about it.
I was playing with Claude most of the morning, maybe a bit TOO much :)
So then I switched over to ChatGPT… not a good option for what I was doing… I wanted a simple email organizer so I don’t have to sort through things, and it did give me a LONG list of stuff I could do, setup filters, install 3rd party applications, create an API for GMail with custom AI filters… But that is a lot of work and I am just playing today!
So… I went to cursor. I was not a fan of cursor several months ago. However, AI is the worst it will ever be today… and apparently cursor has improved so much.
Here’s what I asked:
I want to create a gmail organizer web app. It should ask questions to understand the user, and then connect to their gmail and start organizing based on the survery.
It should not ask more than 5 questions to start.
It should create labels as needed, and automaticlaly categorize emails and move them into those labels and out of the inbox if they are not considered important.
It should return a summary daily of what tasks it completed and how it helped them that day.
It should send a summary of the top priority emails in a succinct way.That’s it. It told me to create a google cloud app, which I know how to do, and an OAuth screen, and get the credentials. And let me not gloss over that. Those are the details that agents can’t do. They can’t work that autonomously yet. There are security risks and well, they just don’t have access to do that. I did play with agent mode on ChatGPT and simply - NO. Claude code is better, but still not 100%. That is the part everyone is missing regarding AI. It can make BEAUTIFUL front-ends that do absolutely nothing. They don’t integrate with 3rd party tools, they don’t create deployment pipelines or scripts, they don’t refactor backend code for performance, unless you ask it to. It does not take the next step that a good engineer would do automatically.
That is the gap we have in AI right now. You see all these fancy demos with beautiful examples of working code (allegedly), they are often simply videos of demos. Ask for the working url, or source code (like I provided at the start). They will come up with excuses at worst, and at best, will come clean about the lies they are telling you about “working code”.
Don’t get me wrong, I’ve done my dog and pony shows, vaporware, and smoke and mirrors demos… But there is no reason any more. We can make it so fast and real, that it is not years to build something, but weeks… sometimes hours.
Ok, back to the code… it generated everything for me. The tailwind config, package.json, components, etc… created a server section for the backend integration for security to GMail, and a full auth scenario.
Nice work Cursor. I’ll let you decide what you think of the UI, but from a single prompt (with one follow up to create a .gitignore file to keep my credentials safe), it worked great.
Not my plan to spend my Saturday like this, but.. hey, wanted to see how fast I can build something. Now I can build that quickly, I want to start doing this more for you. And yes, this newsletter is free, and the code is free, and I’m doing this because I love playing with tech and AI has made it even more fun… when it works. Sometimes it is fun to yell at it and tell it its stupid and it can’t read. But that is more humorous than anything else.
Let me know what you want me to build next and give you the code to play with. Oh, it drives me nuts when people have long youtube videos to make you watch when it could have been a 3 minute video not 30 to show you how to blur something (as my example). Here’s how you blur something on windows. Use edge, open the webpage, press ctrl-shift-s, select the area you want, select the mosaic area, draw over it, and save it. done. That’s it. That did not take a 15 minute video to explain that.
Stop pumping garbage out there to increase engagement, give me real working solutions as fast as you can and I’ll come back for more.
That’s all for today (oh, and this was all written by me real-time, with NO AI (except that top image, I’m not an artist, but with AI, I’m kind of one).
Thanks yall!
Here’s the full workflow:
Then gets some info, for some priority guidance
The name of the app (ZenMail) can be whatever you created in your Google Project, my app is only in test mode and I added my email as authorized only.
Then it redirected to here, and TADA. It pulled in my top emails, I clicked on Organize now, it added labels, removed things from my inbox, and everything worked great and I only have 2 actual important emails to read.
Hope you like this. I was having fun doing it.