A Developer Is Not a Bot, but Won't Mind If You Give Them a Prompt

At MadMountain we respect each other and don't treat one another like machines, but there are some skills we picked up precisely because we talk to those machines. One of them is prompting.

We believe this skill not only lets us work with various AI models every day, but also carries over into our internal (and external) communication.

Trying to imitate AI-speak - “why does this work?”
The ability to write good prompts is about managing information and intent. Although it's associated with typing in dry commands, the mechanisms behind it are identical to those in human communication.

The dynamics of work at a software house often look like this: you drop into a project for a few hours because you're needed. But without onboarding, you won't be able to make good use of that time. So the key is for someone inside the project to give you the complete set of information: your role, the task, the context, the constraints, the expected outcome. Isn't that exactly what AI needs to spit out a satisfactory answer?

Let's not kid ourselves - we will delegate many of these tasks to AI anyway, but there are also ones that a human has to do. And it's much easier to approach a task when you know the basics and the outline.

We've noticed that since we started prompting successfully, our communication has become much more effective. We no longer assume that the person on the other side will figure out what we mean. Instead of “It should be intuitive,” we'll say “The goal of this screen is fast conversion, the user should see the main button above all.”

What are we better at than AI?
When we get a prompt that needs improvement, we know what to ask to clarify the task. We don't jump to generating slop just because the task description is supposedly sufficient. In other words - we think, we react. Just like the human on the other side of the screen - our end user, who doesn't automatically click “Buy now” just because the button is styled appropriately and found the right selector in the DOM tree.

Similarly in communication with the business - it's much easier for us to get the information whose absence is a “blocker” if we ask the right question. Instead of deliberating over complex problems the system has to solve, knowing our roles, context, and constraints, we use Occam's razor, i.e. a single question: do you want a or b? Because usually both outcomes can be achieved, but our approach to the topic, that is, the task, will depend on the answer.

So we know exactly what we're missing in order to start working. We don't waste tokens on yet another pointless iteration of the same issue.

The more we use AI, the more sensitive we are to content it has generated. But it turned out that, thanks to this, we also learned to communicate and collaborate much more effectively.

Looking for a team that uses AI consciously? Write to us or leave a comment 📩