Creating your own playbook for AI engineering

In software, we are used to hearing the term best practices.

Writing tests is a best practice. Version control, CI/CD, code reviews... they are industry standards that save us from rethinking the same decisions over and over again.

I don't know many teams that start a new project by seriously debating whether to use Mercurial instead of Git.

Now, with AI, everyone is trying to find the new best practices.

We hear about RPI (Research, Plan, Implement). Spec-driven development. Context engineering. Harness engineering. Code review agents. AI coding workflows.

But two things are happening.

First: there are no real best practices yet.

We are still creating them. We don't really know what works best. That's why it is so important to look at what other teams are doing, experiment internally, and build your own playbook.

Second: the context is moving faster than our ability to adapt.

What felt like a best practice two months ago may already feel outdated today. A year ago, LangChain was everywhere. Now I barely hear teams talk about it.

And we may be entering another major context shift: AI costs that were heavily subsidized may not stay that way forever.

The reality is that this will keep changing.

Some practices will stay. Some will disappear. Some will only make sense for a few months.

So for now, the best thing we can do is watch what other teams are trying, experiment internally, and avoid turning any AI workflow into dogma too early.

AI engineering is still too young for that.