I started with just ChatGPT for the preliminaries.

I wrote a description, loose set of requirements then had the AI review the requirements for completeness and consistency, then draft them into a requirements.md file.

Then I gave the requirements to the AI and had it generate an architecture and design doc.

Then I had the AI review the design against the requirements a few times, refining them.

I also did a couple of rough paper sketches of layout and photographed them (digital or scan would be fine/better of course).

Only then did I switch the agentic coding AI.

Again, I broke it down into steps.

First, mock up the UI/UX with dummy data.

Implement the backend needed to support the UI.

Hook it all up.

Refine and add features.

Have it do code reviews to remove dead code, consolidate redundancies, etc. Always make sure it meets its requirements and design (modify design if necessary so that it documents the real state). It can also maintain your Readme.md.

I did this in a long afternoon to create a utility site that tracked Azure DevOps PRs across multiple projects and repos, creating links to YouTrack tickets, and providing basic searching and sorting (so a lot of API calls and state persistence, but no local database). I only had a step “fail” once or twice, usually due to misunderstanding or vagueness. No real errors.

By the end of the afternoon, I gave the tool to my team and it’s been being used without error since. a few new features were requested and I did the last few steps for each new features - add feature, code review, requirements review, update design and Readme.md.

I did it as an experiment with my constraint being that I refused to edit code, but I did review the code it made and made it make fixes.

I didn’t do it as a one shot. Roughly, I:

* sketched out a layout on paper and photographed it (very rough) * I made a list of requirements and has the AI review and augment them * I asked ChatGPT outside of the IDE to come up an architecture and guidelines I could give to the agent * I presented all of that info to the AI as project guidelines and requirements * I then created individual tasks and had it complete them one by one. Create a UI with stubbed API calls and fake data, Create the service that talks to AzureDevOps and test it, create my Node service, Hook it all up, Add features and fix bugs.

Result, fairly clean code, very attractive and responsive UI, all requirements met.

My other developers loved and immediately started asking for new features. Each new feature was another agentic task, completed over 1-3 iterations.

So it wasn’t push button automatic, but I wrote 0% of it (code wise) and probably invested 6-8 total hours. My web dev skills are rusty, so I think the same thing would have taken 4-5 days and would not have looked as nice.

I mean, it’ll get better, but it ain’t there yet.

Problem statement: We have four major repos spanning two different Azure DevOps servers/instances/top-level accounts. To check the status of pull requests required a lot of clicks and windows and sometimes re-logging in. So we wanted a dashboard customized to our needs that puts all active pull requests on each repo into a single page, links them to YouTrack, links them to the Azure DevOps pages, auto-refreshes, and flags them by needing attention for approval, merge conflicts, and unresolved comments. And it would use PATs for access that are only stored locally and not in the code or repo.

AI used: I began by describing the project goals to ChatGPT 5 and having it suggest a basic architecture. Then I used the Junie agent in JetBrain’s WebStorm to develop it. I gave it the ChatGPT output and told it to create a Readme and the project guidelines. Then I implemented it step by step (basic page layout, fill with dummy data, add Azure API calls, integrate with YouTrack, add features).

By following this step by step iteration, almost every step was a one-shot success - only once that I remember did it do something “wrong” - but sometimes I caught it being repetitive or inconsistent, so I added a “maximize code reuse and put all configuration in one place” step.

After about 3 hours, some of which was asking it code to my standards or change look and feel, I had a very full featured application. Three different views - the big picture, PRs that need my attention, and active PRs grouped by YouTrack items. I gave it to the team, they loved it and suggested a few new features. Another hour with the Junie Agent and I incorporated all the suggestions. Now we all use it every day.

I purposefully didn’t hand edit a single line of code. I did read the code and suggested improvements, but other than that, I think a user with no programming experience could have done it (particularly if they asked chatGPT on the side, “Now what?”). And it looked a helluva lot better than it would have if I coded it because I’m rusty and lazy.

Overall, it was my biggest success story of AI coding. We’ve been experimenting with AI bug triage, creating utility functions, and adding tests to our primary apps (all .NET Maui) but with a huge code base, it often missing things or makes bad assumptions.

But this level of project was near perfect capability to execution. I don’t know how much my skills helped me manage the project, but I know that I didn’t write the code. And it was kinda fun.

So I bought a whole bag of it and ate a piece every day or so. After a week, I wasn’t cringing as much. After two or three weeks I started craving it. By the end of the month, I liked it. I don’t love it, but I did buy another bag when that one was done. And yes I know the health risks, but I’m never going to be eating a bag or two a day.

The weirdest, though, was cilantro. I’m in the genetic group that thinks it tastes soapy. And yet, after trying it enough, I love it.

I’ve found it useful, but I recommend a “give me an honest critical evaluation as if you were an editor/agent/publisher” and “Is this derivative of anything?”

Anyway, it’s 40 years later and I just read this article and said, “Oh! Now I get it.” A little too late, for Dr. Hippe’s class.

I was a theater projectionist, back when you had 20 minute reels you had to constantly change, while babysitting two high-voltage, water-cooled, carbon arc projectors. Sometimes the film would break and you’d have to splice it. So when the theater got a print in, you had to count and log the number of splices for each reel, then the next theater would do the same and retire the print when it got too spliced up (plus, sometimes if it was the last night of a run, some lazy projectionists would splice it in place with masking tape and then you’d have to fix it). Sometimes you had to splice in new trailers or remove inappropriate ones as well.

Anyway, you counted splices by rapidly winding through the reel with a benchtop motor with a speed control belted to a takeup reel while the source spun freely. Then, while letting the film slide between your fingers, counting each “bump” you felt as it wound through. I was told to ground myself by touching the metal switch plate of the speed control knob with my other hand. One night I forgot and let go until my hair started rising. I’d gone through most of the reel at a very high speed and acquired its charge.

I reached for the switch plate and shot an 8-10” arcing discharge between the plate and my fingers.

Lesson learned, I held the switch plate from then on.

I wonder how well it could do for game summaries and recaps?

I think an interesting challenge would be looking at a board configuration and scoring it on how likely it is to be real - something high ranked chess players can do without much thought (telling a random setup of pieces from a game in progress).

I once commented on Twitter that the Anansi Boys audiobook read by Lenny Henry was better than the book. Neil Gaiman responded, “I agree”.

Then, a year or two ago, I read about how bad the early translations were, so I picked up a new English translation. Wow, what a difference. Now it’s one of my favorites.