AI feels like all the rage lately. When does our company use it?

The determining factor comes down to labor costs over the short and long term, just like any other technology.

If you can describe a logo idea and have it ready for production in 5 minutes, why would you not? That's exactly how our company logo was built at its inception. Thought of the name, prompted for a logo, and it came out pretty good. It originally had a background, cropped that out and made it transparent, called it a day.

The past decade of Lean Startup thinking has had us in the Build-Measure-Learn loop, and building was done in 1-2 week iterations. Now we can pack logo creation into 5 minutes of that iteration and use the time for other things. We can pack days of website development into minutes as well. Great. We haven't even finished our morning coffee yet.

These tools are free for now, so it's extremely appealing. But how many tokens are we using? The cost is subsidized for now. There's a gold rush. What are the AI platforms getting out of it? Training for their models and testing. They're largely burning investor cash.

What happens if we build our business on the model of free tokens, and they dry up and we have to pay the cost? We are out of business and it's our fault for not taking the economics into account.

As a mitigating measure, we use AI as loosely as possible in training or slow periods, and as much as possible on game days. We learn from the AI to the extent possible, and use it as similar as possible to human workers. Software mirrors the communication structure of the organization it exists within.

Rather than prompting big bang full stack solutions, we use our full stack expertise and prompt for concepts that we can rapidly learn from and incorporate into our products and services.

We will also take client preferences into account. We find it hard to reasonably discriminate vs "100% hand crafted" or not, because we treat all code with the same care. We don't use Agents, we use Chat, like talking with a remote worker. We design with it and agree upon the specs, and then we have it implement those specs. LLMs are transformers, that means they transform clear specs into code exceptionally well.

There are some applications of AI that are sensitive and require guarantees like traceability. In those cases we can use Generative AI to crate a system that runs in a fashion that meets and applicable requirements such a regulations. The generated portion can be reviewed by a human, as though a contractor wrote it.

We don't insist upon AI use; we understand human factors like marketing promises. Some people have strong feelings against it, so if we sell a solution that is AI-free we'll ensure that's the case.

However you feel about AI, you can go ahead and Prompt a Human for a Custom IT Solution.