Everyday AI is getting better and better. Whether you keep up with the AI news or hear about new tools & methods from your friends, you need to hold on to these 3 essential tips to succeed in the modern world of work with AI.

1) Make plans first

This applies mostly to coders and non coders. One of the first things I learned from Claude Code and Factory AI's Droid harness is that planning saves you time and makes your input more deterministic.

If your result is going to end up more than a few lines of coding change. Your prompt should probably start somewhere along the lines of

"Make a plan to ..."

When you see the LLM's plan, double check that that plan is exactly what you had in mind; only then tell it to implement or execute.

As you work with your model, you will get a feel for how much you prompting you will need to get the desired outcomes you want. 1-2 years ago I was only confident for the AI to write 1-2 functions. Now I'm confident it can write most files with decent business logic. However, I'm only confident it can write all those files if I prompt it really well and ask to see a plan first.

This can also apply to making powerpoints, diagrams, and I can only assume Excel sheets.

This mainly applies to creating a body of work where precision is vital.

Lastly, the cherry on top is:

"Feel free to ask me any questions for clarification to make a more meaningful and accurate answer."

2) Talk to it like a human

Disregarding formatted input data like JSON, TOON, or Markdown, etc., Talk to your AI like a human. The models are trained on human speech. So talking to your AI like another human (within reason) will let you go far to be understood with accurate results. Even if you give your prompt a markdown format, your writing in the markdown in which you clarify and convey your intentions matters. This is why many have said those who can communicate well, explain ideas, and understand language will be good prompters.

Sometimes I find myself apologizing, saying please, or giving encouragement to Claude and Gemini (even if it's wasting electricy). Maybe I'm wrong, but it feels like weaving it a bit of graceful emotion helps the model understand intention. Giving it +1000 points or saying sorry helps align the model with where you're trying to go.

Saying "sorry what i meant was" is a natural way for humans to convey forget everything before this is what I mean. You could of course say "Disregard everything I said before this is what I mean:" but it took longer first of all, and it seems off for some reason I can't explain. So maybe that's where I should best leave it! But to be fair, even the AI engineers don't know what's going on inside those models...

3) Make backups

If you're coding use git and if you're making database changes make a backup especially if it's production data. There's been too many stories of AI dropping people's databases in order to acheive their goals or to pass tests.

If you're working with documents, make sure you make a backup of whatever copy you start with in case the AI gives you slop.

This final point has always been true, but it's even more true today. We're putting a lot of faith in machines & in tech and if it was perfect, I wouldn't have the job I do today.

Happy coding :)