My First Vibe Coding Hackathon

This past weekend my company did their first vibe coding hackathon. If you're not familiar with vibe coding, it's a new craze where you let AI Agents code your entire application and you just prompt the AI with "vibes." Pure vibe coding requires having zero coding knowledge, but developers have started finding ways to have the AI do most of the code writing while they review and analyze the output and make tweaks when needed.

Over the weekend, we were asked to use Replit and their AI Agent to create a working prototype of a useful application. The top 3 applications would be given money to the company store (which I personally enjoy since there are some quality items there) and we could present our applications to the company leadership. I threw my hat into the ring because why not! I hadn't vibe coded yet, but it would be good to know what's out there and the current state of AI.

This post won't go into the nuts and bolts of the app, but in about 7-8 hours I was able to build a working application that was 90% functional. It was a feature request tool that allowed your customers to request features they'd want to see in internal applications developed at their company. I took this idea actually from Compass Real Estate, where I used to work. They allowed their real estate agents to submit feature request for the Compass software tools. Much like Compass' tool, users of my application could:

• Submit feature requests
• Upvote other people's requests if they liked them (Reddit style)
• Comment on feature requests
• See new requests submitted in the past 48 hours
• Review previous completed requests

Administrators had a special Admin Center page where they could:

• Perform all the tasks in the above list
• Consolidate duplicate feature requests
• Add special "Admin comments"
• Delete feature requests
• Update the development status of feature requests

A screenshot of the main application page is below:

A screenshot of the admin center is below:

What I learned about Vibe Coding

AI Agents are not bad. They are not experts yet, but they do a pretty good job getting a basic application spun up really quick. This whole application took me two days, but really I spent 7-8 hours total doing everything. 

A lot of that 7-8 hours though was time spent waiting on the AI to finish analyzing, coding, reviewing, and processing all the changes I would request. While the AI is pretty good, it still can take 5-10 minutes to make a change, especially as the code base increases. 

Another good chunk of the time was taken up fixing bugs introduced by the AI when working on something completely unrelated. Several times the AI would mess something up in the application that was completely removed from what it was supposed to be working on. For example, when I had it work on the left-side navigation bar of the Admin Center, it also put the Update Status, Consolidate, and Delete button in one inline row and changed the button label for Update Status. Why??? I spent about 20 minutes getting the AI to fix the buttons without hurting anything else.

The AI Agents was not able to get the Search functionality working perfectly. I ended up flipping back and forth between two bugs in the search tool. When I fixed one issue, a second bug would appear. When I fixed that second bug, the first one came back. After a few rounds of that, I got the search working as best I could and left it at that. At the time of this writing, that is an active bug.

Another thing I learned was the AI is very verbose with writing code. This application has to be several thousand lines long. I feel like a human software engineer could write this in half the lines. But one nice thing about Replit's AI Agent at least is that it would give you an outline of what it was doing and give you a hyperlink to go in and check the specific file it is working on. I feel like if I was not rushing and trying to create as much as I can in one weekend, that is a very useful feature in the AI logic.

Overall Thoughts

Overall I felt more like a Product Owner this weekend and not a developer. I certainly didn't personally build this application, but I provided the ideas, creativity, prompts, and analysis that gave the application the look and feel. I wonder if that's what the developer profession will be one day - a world where we will all be product owners of some sort just requesting changes and troubleshooting manually when needed. Or maybe humanity will think "nah, this sucks. We want humans!" I hope it's the second option. 

Right now, the Agentic AI could get me nearly there. But it can't get this application to 100%. Advances are being made, and I'm sure in the future AI will be better and quicker. However for now, you still need some humans around. And I hope coding doesn't lose it's humanity any time soon.

PegaWorld 2025 (and my general thoughts on AI in 2025)

 

PegaWorld was a few days ago, and once again it was a huge blast. I'm fulfilling a promise to myself to actually write about it this year after neglecting to last year. I'll include a few things from 2024's edition though, as this year was sort of a sequel to the events from last year.

AI is here, and even low-code tools (which already do most of the code writing for you) are embracing it to the max. Last year, Pega announced several AI tools coming to their platform and this year followed up with the newest fad in AI - Agentic AI. For a platform that focuses on processes and workflows, I suppose this fits.

The conference itself

The conference itself, once you can get past some of the cheesy humor and jokes during the keynotes, is actually pretty fun. Alan Trefler's talk was a great introduction to the conference, and the point I took away from his talk was that you cannot stuff just any AI into a solution and think you will be okay - you need to use the right AI for the right problems. He compared it to attempting to use Chat GPT to help him with a chess puzzle. Chat GPT could not solve the problem, and instead gave Alan several incorrect/illegal moves in chess. As soon as Alan went to another AI that specialized in chess, it was solved in under two seconds. Thus begins a conversation about Pega having the right AI to solve workflow automation problems.

For the developer though, the keynotes are nice and some of the workshops and breakaway sessions can be really useful. But going to the innovation hub and seeing the new capabilities in action, talking with Pega engineers that built and specialize in that tool, and seeing how it can fit what we do on a day-to-day basis is the best part of the conference. I was able to meet some folks who are trying some new ideas in the Pega community, and I was able to learn about automated testing and network with people who have found creative ways to do it in Pega (a sticking point for our team right now).

I came back with several takeaways from the conference:

1. Become a bigger part of the Pega community
2. Pega Blueprint has changed quite a bit since being announced last year - I need to retrain myself on it and check out the latest updates.
3. I want to learn more about automating testing within Pega via code scripts and accessing JSON files via their API.
4. Pega Insights is a cool tool that we currently under utilize in our app at work - maybe we can use it?
5. There will (hopefully) be a small Pega event in my city later this year, so I look forward to going to that.
6. Pega recently released a whitepaper on AI in Operations, so I've downloaded that and plan to read it in the next week.

My AI thoughts in general

This conference both has me excited for the use of AI in application development and also terrified for the world we are creating. One of the Pega keynotes introduced their newest intern, Iris. Iris is a young girl who started working for Pega this year. She's fully remote, lives in Europe (Pega has a solid presence there, especially in the Netherlands), and has become a top contributor who has helped their employees with admin tasks - the type of stuff interns typically handle.

She's made mistakes on the job, as any person would. But to be frank, she's probably more of a professional and less green than I was when I was an intern. Iris loves her work, she loves where she lives (I believe it was an island with a northern lighthouse in the Netherlands?), and she enjoys working with her colleagues.

And as you've probably already guessed, Iris is also not real. She's completely made up - an artificially intelligent agent. 

On the one hand, this is cool. We've gotten to the point where computers can talk to us and be hyper-realistic. She has an employee picture, Pega employees refer to Iris with the she/her pronouns, and she has a personality that seems to make her human. 

But I cannot help but think we are starting to automate ourselves out of work. Right now, Agentic AI is being billed as the AI that will help you do the tasks you don't want to do. That's nice. But did Iris fill in a gap that would have been filled in the past by a real college intern? How about when Agentic AI gets to the point where it can do entry-level jobs - how long will it be when a CMO or VP of Finance decides that the entry level marketing or finance position can be handled by an Agentic AI instead of hiring a real human to do the same work? Because let's be real: computers have always been faster than humans at tasks that computers are capable of doing.

We're not far from this reality. Agentic AI, and AI in general, has made a lot of strides in the past few years since GPT 3 was announced to the world. And now, we have companies like Artisan in San Fransisco running marketing campaigns to stop hiring humans and instead hire their AI for massive cost savings. (Their marketing campaign is a sad one at that, purposefully created to make people angry and gather likes, clicks, impressions, etc. They should not have been rewarded for that, but alas here we are.)

All that to say, we may be coming to a point where we have to make some decisions with AI. How far are we willing to go to use it? Are we willing to let one AI handle the job of several people for the cost savings and speed? How many jobs are we willing to replace with AI if so? If all companies get on board with this, how quickly does it take us to automate 30% of the workforce out of a job? Will we make decisions today that harm the community at large over a long period of time in order to report good earnings and cost savings for the next quarterly investors call?

You can probably tell where I land on this. 

I was going to write about this in a separate post, but PegaWorld both last year and this year had me leaving with these apprehensions. I didn't come away from these keynotes fully excited for the future we are currently creating. I hope I'm wrong about it, I hope that companies still keep people employed and AI doesn't contribute to the death of several valuable industries and jobs. But after observing the past few years, I get ever so closer to feeling that we're all beginning to hand our jobs over to the control of just a few AI computers.