I was scrolling through my X feed when I saw a gem amongst all the AI news and clickbait. I tend not to get that much out of X these days since everyone is so rabid and FOMO they will post about anything AI-related to try and keep their fickle audiences from turning to someone else for “the news”.

I get it to a certain extent, but then I don’t. Like my last experience scrolling X while eating a salad in front of a window. At a point, I looked at the window and thought, “would it be better for me to just sit and stare at the window vs scrolling on my phone”. I did a quick A/B test and the phone screen went dark. Blissful birds, squirrels, trees, and evertything outside kept my attention and at the end, I wasn’t even angry. I was relieved. So, I tipped my cap to the “free WIFI” sign on the window and went about my day one scroll less frustration.

However, I also tend to get blog post ideas from scrolling X and a particular post caught my eye. I can’t find the complete post now, so I will paraphrase…but it started with “Don’t…” and then listed out a bunch of things the poster thought software developers shouldn’t do. One line in particular almost make salad come out of my nose!

Don’t start every coding session with a prompt

They were referring to using AI assistants/agents in code editors and ostensibly thought it bad practice to use them every time someone reached to modify or create new code.

But this is an incredibly dumb statement, so probably better I don’t have a post to link to, lest the author sink their head in shame.

What is a prompt?

To explore how stupid the assertion “Don’t start every coding session with a prompt” is, we first need to define what a prompt is and what it is not.

A prompt is not simply what you type in a chatbox to get an LLM (Large Language Model) to update code. Technically it falls into that category, but zooming in that narrowly misses the larger picture.

If we boil down a prompt to its most basic level, what we have is simply natural language. In fact, what we have looks eerily similar to what one human might say to another human they manage and want to complete certain work. So in other words a prompt is a “request”.

At any job I’ve ever had, when the boss wants me to do something, they give me a prompt and I reply after considering the context with code or questions I need to further define the task.

So another way to put it is this: a prompt is a ticket.

A Prompt is a Ticket

Yes, a prompt is the exact same thing as a ticket you’d get assigned if you work in a dev team. In fact, I’ve been writing the exact same types of tickets to work with AI agents as I did in my last dev job working with humans.

I quite prefer the AI to the human since it pretty much agrees with any formatting or other requests I have. No fighting over tabs and spaces with Claude. He just gets me. He gets me, I swear.

So was there ever a time where your whole dev team just went rogue, didn’t pay attention to the ticketing system, and did whatever they wanted? Even if you say “yeah, that happened”, how long did it last?

Starting coding without a prompting sessions is incredibly dumb, but it doesn’t surprise me some dumb developer would post such a thing on a social media platform. Not after scrolling X and preferring watching squirrels…squirrels, man.

The Rise of the AI Project Manager

Not only do I encourage you to definitely engage with AI before writing any code, I think you ought to create your own AI project manager like I have done.

Every night, I kick off a standup plan workflow where Claude Desktop reads my calendar and notes via MCP servers and then helps me review what I got done today and what I will do tomorrow.

Then, in the morning I run a standup start workflow to kick off my day. It feels very similar to working in a dev team, something I’ve missed since quiting my last job, and I’d say it is way more effective since the AI never forgets to do something you tell it.

In my experience, humans will almost always forget to do things you tell them. Not because they “forget”, but because motivating a whole team to improve is hard, especially when they see someone sticking to the rules as an “over achiever” they kind of despise.

Hence, this is why most retrospectives I’ve taken part in end with “great idea, let’s see how that works out” and the next retro is “yeah, we didn’t implement the idea. Oh well, let’s delete it from the todo list”.

That will never happen with my project manager, Claude. He can’t physically hurt me, but he can remind me to slap the shit out of myself if I don’t achieve my daily goals. I know. I just had a conversation with Claude about this slap happy topic. I doubt any human project manager would do this for me.