Will you still be building PowerPoint decks from scratch in 2027?
Maybe a few Excel spreadsheets, the occasional Word document, and, of course, some emails?
Maybe you want to keep a personal touch. Maybe you need something very specific. Maybe building it is how you think it through.
Fair. But otherwise, there's very little stopping AI from drafting almost any deliverable, and each new model makes that more true.
So what are we to do with all this extra free time?
Well, unfortunately, there's a catch.
AI rarely gets it right on the first pass, and sometimes it's way off.
Worse, it can be convincingly wrong, or so long that checking every word is impractical.
There are three main ways to address this:
➡️ Let AI check its own work. Great where results can be tested, like code. Harder everywhere else.
➡️ Make human review easier: visual summaries, AI quizzing the reviewer. I think this becomes a big field over the next few years.
➡️ Give AI the right context.
The first two fix the output after it exists. The third fixes it BEFORE.
That's where the biggest gain is.
In my experience, with the right context, the first draft goes from about 60% right to 90–95%, and what's left is final tweaks and validation.
Time spent on context is time well spent, and it's the start of a powerful virtuous cycle:
AI drafts the deliverable
↳ Time saved
↳ Saved time is used to think through problems and enhance AI context
↳ Better context means better drafts and less to validate
↳ More time saved, better outputs and so on…
The best part? This means more time spent on interesting core work and less on the drudgery of drafting deliverables from scratch.
I see this as leading to new roles in organizations. More on this in my next post.