0:02 Working with AI comes in roughly three
0:04 levels. First, we have people who are AI
0:06 curious. This group relies on the free
0:09 tier of AI tools and only uses chatbots
0:10 when someone reminds them to or when
0:13 they're stuck. Level two, we have the AI
0:15 literate. These people pay for AI,
0:17 maintain a prompts database, and they
0:19 know when to use which AI feature
0:22 [music] and model. Level three is AI
0:24 native. And in a nutshell, these people
0:27 have redesigned their workflows assuming
0:29 an AI collaborator exists. Most
0:31 professionals are stuck at level two.
0:32 So, in this video, I'll share the
0:34 specific strategies ordered from simple
0:36 to advanced that will get you to level
0:38 three. Let's get started. Kicking things
0:40 off with the easiest habit to adopt,
0:42 leave AI breadcrumbs. What this means is
0:44 instead of treating AI chats as
0:46 disposable one-off threads that become
0:49 almost impossible to find again, you
0:50 want to create a hyperlink to the
0:53 conversation and paste it directly into
0:55 the document where you're actually using
0:57 the output. I know this habit sounds
0:59 silly and insignificant, but it works
1:00 great thanks to the core productivity
1:02 principle of always organizing your
1:05 information by where you will use it,
1:07 not where you found it. Diving into a
1:08 real world example, here's the actual
1:10 Google Doc I used to prepare for a
1:13 recent work presentation. The final
1:15 outline tab has all my content and the
1:18 helpful hints tab has hyperlinks to my
1:21 AI conversations. Taking a step back,
1:22 let's say I'm building this presentation
1:24 from scratch. I would first ask AI to
1:26 rewrite my initial rough prompt so that
1:28 it's optimized for the model that I'm
1:30 using. And once I press enter, you'll
1:33 notice the URL transforms into a unique
1:35 link. Right? And this is where I'd press
1:37 command or control L to select the
1:40 entire URL, copy, come back to the
1:44 Google doc, type out Gemini, command or
1:46 [music] control K to hyperlink, and
1:48 paste that link. I then copy the
1:50 optimize prompt and paste that into a
1:52 new chat. Make adjustments as needed.
1:54 then go back and forth with the AI to
1:56 brainstorm and refine my presentation
1:58 outline. And of course, I save this new
1:59 chat link in the Google doc as well, so
2:01 I can easily pick up where I left off a
2:05 day or even a week later. Pro tip, add
2:07 context next to each hyperlink so you
2:09 remember why it matters. For example,
2:11 here, this Gemini conversation was
2:14 helping me brainstorm my outline. This
2:15 one was around applying storytelling
2:18 principles to that final outline. And I
2:21 use Claw to refine my final talking
2:22 points. And just to be clear, I like to
2:24 test multiple frontier models because I
2:26 do this for a living. Most people should
2:29 just pick one AI chatbot and get really
2:30 good at it. Here's another real world
2:32 example. Whenever I create a new project
2:35 page in Notion here under the op center
2:37 section, I would add links to the
2:39 corresponding CHP and claude projects so
2:41 I can jump into those AI workspaces
2:43 immediately. Put simply, leaving AI
2:45 breadcrumbs means organizing your AI
2:49 chats by work context and not by date or
2:51 chronology. This beats trying to search
2:53 for that one specific thread from days
2:54 or weeks ago. And the rule of thumb here
2:56 is simple. If the AI conversation took
2:58 more than 10 minutes or produce
3:00 something you'll reference again, anchor
3:01 it to your workspace immediately. [music]
3:02 [music]
3:03 Speaking of systems, I'm actually
3:05 building an entire course on evergreen
3:07 AI skills that teaches universal
3:09 principles for any platform. So, if you
3:10 want a framework that never goes
3:12 obsolete, click the link below to join
3:14 the weight list. Moving on to habit
3:15 number two that requires a bit more
3:18 effort, build an AI swipe file system.
3:20 In a nutshell, instead of prompting AI
3:22 with basic instructions like write a
3:23 business proposal, you provide a
3:25 specific example from your curated
3:29 library aka your swipe file and ask the
3:31 AI to first analyze what makes it so
3:33 effective. Then apply those patterns to
3:35 your new content. For instance, let's
3:37 say you work in OpenAI and you have this
3:39 brilliant idea that you know users
3:42 around the world will absolutely love
3:44 pumping chat full of ads. Instead of
3:46 starting a proposal from scratch, you
3:48 open up your AI swipe folder to find
3:50 examples of business proposals you've
3:52 previously saved. Share them with the AI
3:54 and say, "Analyze the business proposals
3:56 I've attached, list the key patterns in
3:58 structure and tone, then apply those
4:00 patterns to my content below." And you
4:02 paste your advertising product idea
4:04 designed to maximize shareholder value.
4:05 Jokes aside, I guarantee you that
4:07 initial output will [music] be stronger
4:09 than any initial draft you could have
4:11 come up with yourself, not to mention
4:13 the massive time savings. Funny story,
4:14 when I was at Google, I used this
4:16 technique for all my important
4:18 presentations by uploading slide decks
4:20 from Mackenzie Bane and BCG. And senior
4:22 leaders who came from those firms would
4:25 ask if I had also worked there before.
4:28 And I'd be like, "What? No, these
4:30 frameworks and principles come so
4:32 naturally to me. Does it not come
4:34 naturally to you?" And that's why I'm
4:36 not at Google anymore. No, just kidding.
4:38 But as you can see, this technique is so
4:41 effective because it gives the AI a
4:44 clear picture of what good looks like,
4:46 allowing Chachi or Gemini to produce
4:48 output that matches those standards
4:51 instead of generic slot. To close the
4:52 loop, the actual habit you want to
4:54 develop is whenever you encounter
4:56 excellent work in your field,
4:58 immediately save it to your swipe file
4:59 system so you can reference it the next
5:02 time you face a similar task. Pro tip:
5:05 start narrow and expand gradually. Begin
5:07 with just two to three use cases you do
5:08 repeatedly like presentations, emails,
5:11 or reports. And organize your folders by
5:14 use case, not by source or date. By the
5:16 way, this is also the first step in
5:18 making your Google Drive AI ready, which
5:20 is something I dive deeper into in the
5:21 course I just mentioned. And number
5:24 three, we have AI first task planning.
5:25 Heads up, this habit is probably the
5:28 hardest to maintain consistently, but I
5:29 [snorts] promise you, just like going to
5:31 the gym, it will make a massive impact
5:33 over the long run. Put simply, this
5:36 habit involves planning your AI use
5:38 before you start a big piece of work.
5:40 This means breaking down complex
5:42 projects into small concrete tasks, then
5:45 marking the ones AI can and should help
5:47 with. Diving right into real world
5:49 example, I used to be responsible for
5:51 sending these uh weekly newsletters to
5:53 our Google Ads customers uh with the
5:55 goal of driving adoption of new product
5:57 features, aka getting them to spend more
5:58 money. Basically, before writing
6:00 anything, I'd break down the work into
6:03 steps and microtasks. Then I decide
6:05 whether to do each microtask manually or
6:09 use AI. If I use AI, I specify the exact
6:12 tool that's best suited for that task.
6:13 In this case, there are three main
6:15 steps. Step one is to clarify the goal
6:18 and audience. Step two is to draft the
6:19 newsletter. And step three is to refine
6:21 the copy for Google's brand voice. [music]
6:22 [music]
6:25 Now, step 1.1 would be to brain dump key
6:27 information like what the new feature
6:29 is, the benefits, who should use it, and
6:31 so on. This task is manual because there
6:33 are details I know that AI doesn't have
6:36 access to, and I also want to inject my
6:38 point of view. Step 1.2 is to fact check
6:40 my notes from the previous step. Here,
6:42 it makes sense to use AI, specifically
6:44 Notebook LM, since it has the lowest
6:46 hallucination rates. And so I upload my
6:48 brain dump and source documents onto a
6:50 notebook to verify rollout dates,
6:53 feature names, policy details, etc. Step
6:56 1.3 is to turn those fact check notes
6:58 into a structured brief. This is
7:00 obviously also a perfect task for AI,
7:01 but this time I use the standalone
7:04 Gemini app instead of notebook LM
7:06 because the Gemini app is much better at
7:07 creative writing. In the interest of
7:09 time, I'm going to skip over steps two
7:11 and three, but the process is exactly
7:14 the same. List out all the microtasks.
7:16 decide if AI should help and if yes,
7:18 pick the right tool for that specific
7:19 job. At this point, the benefits should
7:21 be pretty clear. First, you cut decision
7:23 fatigue and context switching because AI
7:27 usage is already preddecided at the task
7:29 level. Second, you increase quality and
7:31 speed by matching the right AI tool to
7:32 the right kind of work instead of
7:34 forcing one tool to do everything. And
7:35 the rule of thumb here is for any
7:37 project that will take more than an
7:39 hour, spend 5 to 10 minutes mapping the
7:42 steps and tagging which ones are AI or
7:44 manual. For the productivity nerds out
7:45 there, this is a classic example of
7:48 sharpening the axe, where spending a few
7:50 minutes on planning up front saves hours
7:52 [music] of work later. Pro tip, create
7:54 templates for recording workflows, like
7:56 what I have here for my weekly
7:58 newsletter, so that next time you can
8:00 focus on executing instead of having to
8:02 plan from scratch. This brings us to a
8:04 bonus habit that ties everything
8:06 together. Maintain a prompts database.
8:07 I've talked about this many times
8:09 before. Whenever you write a prompt that
8:11 works well, save it to a central library
8:13 organized by use case so you can reuse
8:15 that prompt whenever you face that task
8:16 again. The worst feeling is writing a
8:18 perfect prompt 3 weeks ago that
8:20 generated a perfect output. But today
8:22 you can't find it. So you try to rewrite
8:24 it from memory and the result is just
8:26 eh, which is also my girlfriend's
8:27 reaction shortly after we started
8:29 dating. Huh. A lot of you have been
8:31 asking for my go-to prompts. So, I spent
8:33 quite a bit of time putting together a
8:35 set of essential prompts that everyone
8:37 can benefit from because I genuinely
8:39 believe you don't need a thousand random
8:41 prompts. You need 10 to 15 battle tested
8:43 ones that you can use [music] every day.
8:44 I've been using and refining these ever
8:46 since Chachi PT first launched. So, if
8:48 you want to skip the trial and error,
8:50 I'll leave a link to this down below.
8:51 See you on the next video. In the
