Why I stopped writing exacting prompts

One of the core beliefs I’ve carried through life is that you can see anything in either a positive light, or a negative shadow. Any idea, topic, philosophical musing, or even the weather outside. A day full of clouds could mean rain, canceled plans and boredom. It could also mean respite from a vicious heatwave and a chance to master that sourdough recipe you’ve been trying to figure out.

Over the last few months I’ve seen numerous posts lamenting the “death of the software engineer”, and recent events at a Rails conference fanned those flames, as they often do, proclaiming that “the time of writing code by hand is at an end”.

You can see that idea in two ways, and I think we should try.

Writing code is fun, and it’s what I’ve done for my entire career

It really does pain me to think that I 1) won’t write code again, for the most part and 2) really shouldn’t write code again because I know how to make Claude do my bidding. It’s more consistent than I am, writes better, clearer tests and can fix its own bugs easily. It can also review its own code better than I can because of its machine rigor.

So what the heck am I even doing here? Pushing buttons and writing prompts, hoping for the best?

I should probably just do an early retirement and move to Portugal.

Shipping apps is what I do, and my career has never been better

I have spent decades writing code and building apps for my clients, including myself, and I wouldn’t say I’m amazing at it, but I got the job done and I enabled hundreds of businesses to realize their vision. Six of those were my own!

I’ve run teams of programmers and I can’t say I enjoy it, but nowadays they do exactly what I say to do (for the most part) and I pay them very little ($200/mo for the lot of them). You wouldn’t know it, however, because sure as heckfire they write code fast and do a really good job.

I have shipped more apps to production in this last year than in the last 5 combined! My sidehustles are starting to see the light of day, and even the goofy little dreams I’ve had (like building something better than Ghost) are starting to take shape.

I love shipping things - I’m good at it! I’m finally building magnificent applications because I have the time, and the desire.

And it doesn’t suck

I feel like the Great Crap Code Debate is a solved thing these days, because these models are so good. They follow instructions well, and yes, you can just shotgun your prompt and hope for the best, or you can use some rigor, spend a few extra tokens, and get code that doesn’t build your technical debt.

I do know that there are plenty of people out there just winging it, but I like to think they’ll figure it out and correct course soon enough. Either way: I know what I can do, which is the point of this whole thing anyway.

To that end: a few weeks ago I trimmed up my Claude Code Playbook, which has the skills and process that I like to use to avoid crappy crap. One of the things I didn’t like was the sheer amount of questions it asked, just to get a sprint out the door. I found myself creating a “rob” skill that answered all of those questions with my preferences (for the most part).

It was at that point that I decided to flex what many people are realizing these days:

Stop telling Claude what to do, and let it do its thing.

Fable and Opus (apply this to any other frontier model, really) are so, so good at “lateral” thinking that it’s just not necessary to be so exacting with your prompts. I’m using Wisprflow (use that link for a free month) for all my prompts these days, and rambling things out, letting Fable or Opus figure out my intent. 95% of the time, they do.

In fact, it’s better than anything I could have done

Part of the “trimming up” process was adding a Product Owner that is focused on the customer experience. Rather than have that at the end, as a gate or acceptance thing, I wanted to add it in the middle, where it can expand the “delight” a customer would feel using my stuff.

I decided to base the skill on Steve Jobs, and to place the skill in the design phase, working with the Architect to create something wonderful.

Here’s the prompt I used:

One of the things I'm missing here is a product owner: somebody who can see a vision and who might step outside of the spec just to build something magnificent based on their own inspiration and understanding of the human condition. I always think of Steve Jobs. He was the ultimate product designer in my opinion.  
  
With every single sprint, with every single effort, I want the product designer to be woven in to automatically add and round out a feature set. To do things perhaps that weren't asked for, but not by a lot, just by a little. Something that would make someone say, "Whoa, I didn't ask for that but I LOVE IT" I want the answer to be "because it's because it was obvious it needed to happen to make the end user, the customer, absolutely love the app". 

The skill that Claude created is now in my updated Playbook, which you can get right here. I also made it much, much easier to build up a sprint and get rolling, without a bunch of questions.

I used to have a skill called rob-tech that would answer all of the questions from design and explore because sometimes I just want something to be done and I feel like its pretty obvious based on the existing architecture and my previous answers.

That’s obviously not a good idea. Instead, what I need, is a skill with “vision”, which I know might sound a bit silly when discussing AI, but if you ask Fable to help you round out a spec or “add some much needed love to the experience”, it’s quite good at it!

I know this because that’s precisely what I’ve been doing with this little list app I’m using right now. It’s so much fun to think of a feature and add it, but when it comes back with the product-designer's additions… well let’s just say it’s unreal.

This process led me to consider doing a “pro” version in much the same way Ghost has both a pro and open source version. In fact: my goal here is to build the Ghost platform I’ve always wanted, and then some.

Before I go further on that, yes, I’ve tried to help the team through offering suggestions, bug fixes, and feature requests. Some PRs made it in, others were closed off fast. I even spent 90 minutes on a call with Hannah Wolfe on Thanksgiving a few years back, hoping to help in more structured way.

I’ll just summarize my interactions this way: they know what they want to do, and when they want to do it. They’re pretty forthright about sticking to their vision, and I say rock on! Me too - which is why I decided to create my own.

Anyway: my pro version is a port of the open source version that runs on Elixir/Phoenix and has quite a few more bells and whistles to it, specifically because of my product-design skill. So many things I just didn’t think of were added automatically without my having to ask for them. A few, of course, weren’t wanted but I told Claude to argue with me and fight for their existence… which is a whole other post.

It just keeps being fun

Last week I was chatting with Aaron Wislang, telling him about my goofball project and he asked about theming. This list server has morphed into a personal blog/content engine so, of course, it needs a theme! 45 minutes after the call and some work with /high-end-visual-design, Kolea (this app) now has themes!

Here’s one of them, called “Night Drive”:

It also has instructions and skills for working with those themes so you don’t need to edit anything by hand.

My next task is to consider how far (and in what shape) I want to take paid content, if at all. Comments too, and even some community elements. Why not?

It’s just tokens, and we can spin gold from them.

Doin it live

If you’re interested in learning about this in a live, cohort setting, there’s still some room in the Fall Cohort, which you can read about here. One of the suggestions that came in is a great one:

I would love to learn how to take a side hustle or passion project and turn it into a paying business.

Oh hell yes! Kolea started out as me wanting to save a few bucks and to see what’s possible with Opus 5.5. It moved from that to “I NEED THIS RIGHT NOW” in a matter of 10 days. It took another week to sketch out and begin the build of the pro version, which I think will offer some very, very compelling features at an absolute fraction of the cost of Kit, Drip, Mailchimp, and others.

Along the way, I’m learning all about deliverability, DNS, async bulk processing, and more.

This is, truly, the most fun I’ve had in my career!

So, tell me: what are you working on currently? What’s your side hustle or passion project, and do you see yourself making a living from it?

Thanks for reading!