The genie is out, for good or bad

But to serve you well, you must feed it so it does not get mad

It understands it all, but text is best and it better be your own

All hands on keyboards, everything must be written down


With the rise of generative AI, I believe writing is more important than ever.

If you want the genie to work for you, you have to give him something to eat. And what better form than text?

For sure, LLMs are multimodal and can do wonders to process images or videos. But simple text is an efficient way to convey information. Even more, when counting tokens and staying below quotas is all the rage, you want to be frugal.

I have already claimed my preference for written communication but I think now is the time to document everything, so that knowledge is not lost and can actually be leveraged.

In practice, we are looking at:

  • documenting how you work, your workflows, the processes you have not yet automated with tools
  • keeping track of progress, work items, achievements in a ticketing system or simple notes
  • writing down designs, thoughts, roadmaps, anything that can later morph into bits of specifications to input to your coding agent.

All of that, using a textual format. Markdown is very popular for that. Use textual formats for diagrams, sketches too, ascii art is on the rise! PlantUML, Mermaid are good tools.

Remember to work in the open too, in a common company setting, there are very little reasons to hide things from your colleagues.

What we want to avoid is:

  • coffee machine discussions that leave no trace
  • meetings without notes
  • one-to-one conversations that are not shared
  • overuse of videos, whenever I see a bug report that simply consists of a video, a voice inside of me screams!!! As much as a video can help for a UI issue, providing context and reproducible steps is even more useful, in terms of discoverability and future analysis.

It does not mean we should forbid all human interactions, but think how much information is shared every day and how to give it some persistence.

A final note on privacy. The above applies in a professional context, where vetted AI tools are available. When you trust your providers, and hope they will stick to their contractual clauses. When data can be safely shared. You might want to thread differently with your personal data.