It’s maybe a question of experience. I’ve been using my new coding agent for some months now. And I’m just getting into the flow. I know enough about the ruts and potholes to avoid the worst of them.

We’re a small software company that creates custom solutions for businesses with a unique approach to product. Because we’re so deeply embedded in the businesses of our clients, we not only build, maintain and support, but we also initiate.

One of these initiatives is a reporting tool that we use to keep our clients up to date on what we do for them. They might have a long term commitment contract, but we feel we should tell them exactly how much effort we spend on their business.

So every week they get a nicely formatted email with a list of tickets we’ve worked on, where we have AI reformat tech jargon into understandable sentences. Generated by a tool we built ourselves. Because, hey, we’re a software crew.

But the reporting tool we built has a flaw.

For the uninitiated, it’s absolutely impossible to guess how to use it to send the messages to our clients. It’s hidden away among multiple screens that seemingly have nothing to do with the task at hand.

Mostly not a problem, because our people are of course NOT uninitiated, they have the rest of us there to help.

Still, for me, it’s a continuous source of irritation. Every time I look at it I want to fix it, but no, say my colleagues, there are 100 more pressing things.

But yesterday I couldn’t contain myself any longer. It was the end of the day, the so-called more pressing things were done. I should’ve gone home but didn’t.

I asked My AI.

🕕 18:15: (first prompt, where I describe the feature, where it’s currently located, and that I want the interface on the dashboard)

🕕 18:17: (second prompt, where I explain why we need an asynchronous task and a progress bar)

🕕 18:21: (third prompt, where I ask to improve the placing of the new interface)

🕕 18:23: (fourth prompt, where I test for the first time and discover an issue)

🕕 18:25: (fifth prompt, after I discover a second issue)

🕕 18:27: (sixth prompt, after I come across a third issue)

🕡 18:30: works.

We were done in 15 minutes.

And then I wanted it a little bit nicer, and we got stuck in that frustrating loop that always seems to appear when you have done what you set out to do but want to make it a little bit better. Which cost another 15 minutes.

Commit and push.

Done.

I no longer make assumptions about how much faster working with coding agents is than coding by hand. Ten times faster? Fifty times? A hundred?

The true relevance of using a coding agent is not speed.

It’s that you can take on tasks on a whim in the confidence that you can complete them while others have lunch.

Many of my colleagues share this feeling. They see something to fix, they improve because they can. We should all do that.

Imagine how much more beautiful the world would be. 🌅