0:03 How can you get more out of Claude Code
0:07 and Claude Opus 4.5? Well, I've got good
0:09 news from you. Anthropic has actually
0:11 over the last 12 months have been
0:14 posting in their in their docs, blog
0:17 posts, kind of teasing on X about how
0:20 you can prompt these products to really
0:22 get the most out of it. But the thing
0:24 is, people haven't put it in a full
0:28 guide. So, I did the hard work to make
0:30 it easy on everyone. By the end of this
0:32 episode, you will learn 10 techniques
0:35 for how to prompt Claude to get the most
0:37 out of it. Super simple techniques
0:38 anyone can learn. I'm going to show you
0:41 real examples, easy to understand, and
0:44 [music] frameworks to help you crush it
0:47 with Claude code and Opus 4.5. Let's get
0:58 >> So, the first tip is I know this is
0:59 going to this is going to upset a few
1:02 people, but the tone of collaboration is
1:04 really important. You're going to want a
1:06 friendly and clear and firm tone because
1:08 that yields better results and more
1:11 direct results. So, what's an example? A
1:13 vague request might be something like
1:16 fix this grammar in this now. You know,
1:19 but the problem with that is, you know,
1:22 o it leads to overly cautious pre-anned
1:24 or basically just less helpful responses
1:27 as the model tries to deescalate.
1:29 Politeness can sometimes result in
1:32 chatty, less direct answers. Now, if you
1:34 do, you know, an architected brief, and
1:36 this is what the folks at Anthropic
1:39 suggest you do, do something like,
1:41 "Please review the following text for
1:43 grammatical errors and suggest
1:46 corrections. My goal is to make it sound
1:48 more professional and confident. This is
1:51 direct, this is respectful, and it
1:53 provides context, which is what
1:55 anthropic needs in order to get you the
1:59 result, you know, that you want. So,
2:01 really important. I know some of us are
2:03 just kind of mean to our LLMs. I've been
2:05 there, you know, but treat it like a
2:06 teammate, right? You would never want to
2:09 be mean to a teammate, uh, especially if
2:12 you want to get them to produce. So uh
2:15 rule rule one of 10 uh is the tone of
2:18 collaboration. Rule two of 10 is the
2:22 principle of explicit explicit explicitness.
2:24 explicitness.
2:26 So state your request as a clear
2:28 actionoriented command with all the
2:31 necessary details. So and I used to do
2:33 this actually. I would I would do like a
2:35 vague request like I need a bunch of
2:37 blog post ideas. But the problem is it's
2:40 passive. It's not specific. And then you
2:43 just get this generic AI slop.
2:45 Architected brief. What's the
2:47 difference? Okay. Generate 10 blog post
2:50 titles about the impact of remote work
2:53 on urban planning. The title should be
2:55 engaging for an audience of city
2:57 officials and real estate developers.
2:59 This prompt uses an action verb.
3:01 Generate. You're going to you're going
3:03 to want to, you know, use action verbs a
3:06 lot. It specifies the quantity 10 and
3:08 target audience. Those are the three
3:10 things that you're going to need. gener
3:13 action verb uh the quantity and a target
3:16 audience. Every highlighted phrase adds
3:19 a layer of useful constraint. This works
3:24 extremely well. Three on 10. A well the
3:27 rule here is a well-defined box produces
3:29 a more creative result than an empty
3:31 field. So a vague request would be
3:33 something like write a short story about
3:36 a detective in in the future. Problem is
3:38 the possibilities are infinite and that
3:41 leads to cliche AI slop unfocused
3:43 output. Architected brief. What's the
3:45 difference? Write a short story no more
3:48 than 500 words in the style of Raymond
3:51 Chandler. You can even do like in the
3:54 style of Ernest Hemingway meets Raymond
3:56 Chandler. Or you can even put three or
3:58 four or five different people. The story
4:00 must feature a robot detective
4:02 investigating a data theft on Mars. Do
4:05 not use the word cyber. So, you've added
4:07 constraints on length, constraints on
4:09 style, constraints on character,
4:12 constraints on settings, and even
4:15 specific words to force the AI AI into
4:18 more creative and specific solution. I
4:20 know this takes more time. I know that
4:22 you know it's kind of measure twice and
4:24 cut once, but you will get more if you
4:27 define the B boundaries.
4:30 Number four, the rule is draft, plan,
4:33 then act. Use the AI to generate an
4:36 outline or roughed version first. So,
4:38 don't try to get a perfect fi final
4:40 product in one go. I know we all want to
4:42 one prompt it and just it feels so good
4:45 to one prompt it. I've been there. But,
4:47 uh, you know, the reality is working
4:50 with the AI to create and then refine a
4:54 plan or outline is a way more reliable
4:56 path to a high quality result, and
4:58 that's what we're out for. This lets you
5:00 course correct early. So, think of it
5:02 like this. The initial prompt might be,
5:04 I need to write a report on the benefits
5:08 of a 4day work week. Ask for a plan.
5:10 Step one, first propose an outline for
5:13 this report. Then refine the plan. Step
5:15 two, that's a good start. In section
5:18 two, please add a subpoint about
5:21 employee retention. Step three, uh you
5:23 can do you step three is the execution
5:25 layer. Excellent. Now, write the full
5:28 report based on this revised outline. So
5:30 you you know it takes longer, but the
5:33 the al it takes longer. It feels like it
5:34 takes longer, but it's actually going to
5:36 save you time because you're going to
5:37 get a better outcome and you're not
5:39 going to have to reprompt and reprompt
5:41 and reprompt. If you have a business
5:44 that's doing at least $50,000 a month in
5:46 revenue, I've got something interesting
5:48 for you. It's called offline mode. It's
5:50 a 2-day event that me and my team are
5:53 putting on at a 20,000 plus foot square
5:55 foot mansion. Yes, this is what it looks
5:58 like on January 23rd and January 24th in
6:00 Fort Lauderdale, Florida. I'll include a
6:02 link in the description if you're
6:04 interested in coming. But it's basically
6:06 for people who have a business that's
6:08 kind of cranking, but they really want
6:11 to put it in rocket ship mode. They want
6:13 to create a set of businesses that
6:16 generate tons of money, tons of cash
6:18 flow, tons of product market fit, tons
6:20 of impact. Um, but they're not just
6:22 quite there yet. It's also about, you
6:25 know, making your business AI first. How
6:27 you can actually, you know, build not
6:29 just one product, but multiple products.
6:30 And you're going to leave with, you
6:33 know, tactical, uh, answers to your
6:36 questions. So, um, if that's you and
6:38 this sounds interesting, uh, I'll see
6:42 you there. Rule rule number five, uh, is
6:45 demand structured output. The AI is
6:48 fluent in many formats beyond pros. So,
6:49 you know, a vague request might be
6:52 something like, "List Apollo missions
6:54 and some facts about them." So, you're
6:56 going to it's it's it's it's basically
6:58 going to give you a simple unstructured
7:00 paragraph. And that's going to be hard
7:02 to parse. Now, an architected brief
7:04 might be something like, "Provide the
7:07 list of the last three Apollo missions,
7:09 15, 16, and 17. For each mission,
7:11 include the launch date, the crew
7:14 members, and a key specific achievement.
7:15 Present this information in a markdown
7:17 formatted table." So what you might have
7:19 to do is go to something like perplexity
7:23 and do uh you know a prompt there to
7:25 understand a little more about what you
7:27 want an architected brief to look like.
7:29 Um but you're going to end up getting
7:32 you know a way better output here. Like
7:34 you can just see this markdown file if
7:38 you're on YouTube. Um also like and
7:39 comment if you're liking this sort of
7:42 stuff. I'll do more of it. um requesting
7:44 a you know so look at this markdown you
7:47 know it's just way way better than what
7:50 you'd get if you if you did uh did did
7:52 something else and just went like a
7:55 simple vague sort of thing. So might
7:56 take you more time a little bit
7:58 initially but again you're going to get
8:02 better output six explaining the why. So
8:05 the golden rule here is the explaining
8:08 the why behind an instruction helps the
8:11 AI understand your true intent. So
8:13 instead of saying give me five marketing
8:16 slogans for a brand new coffee where the
8:18 AI basically has no context. It doesn't
8:19 know the brand values. It doesn't know
8:21 your audience, your community who you're
8:23 going after, it doesn't know your unique
8:24 selling proposition, do something like
8:27 this. Give me five marketing slogans for
8:30 a new brand of coffee. The key is that
8:31 our beans are ethically sourced from
8:33 small independent farms and our target
8:35 audience really important you put this
8:38 in here is environmentally conscious
8:41 millennials. By the way,
8:42 it's not just that you put our target
8:44 audience is millennials. You have to you
8:47 have to define it even even more. Go
8:50 niche and then go more even nicher.
8:52 You'll get better results. The slogan
8:54 should reflect quality and
8:56 sustainability. So by providing the why
8:58 basically ethically sourced for
9:00 conscious millennials the AI can
9:01 generate far more relevant and targeted
9:04 slogans then you get a better getting
9:08 better results. Seven on 10
9:11 the art of brevity and verbosity.
9:14 So the rule here is explicitly command
9:17 the AI to be more or less verbose to
9:20 match your needs. So you're in control
9:23 of the output length. Use simple direct
9:26 phrases to guide the AI. So let's just
9:29 show some some different examples. So
9:31 maybe you want the expert here. Explain
9:33 photosynthesis in detail for a college
9:36 biology student. Think step by step to
9:39 ensure accuracy. Boom. Look at look at
9:43 how uh how expert this looks like. I
9:45 mean it's throw it's throwing me back to
9:47 you know my chemist my biology and
9:49 chemistry days just seeing stuff like
9:52 this when I was in school.
9:53 Um, maybe you want something that's
9:55 brief. Explain photosynthesis. Be
9:57 concise and use bullet points. Sometimes
9:59 here I'll say like explain like I'm
10:02 five. Explain like I'm 13. Explain like
10:04 I'm 17. Um, and then you can see the
10:06 bullet points. And then here, oh here
10:09 the simplifier. Explain photosense like
10:12 like I'm 5 years old. So these are three
10:14 options. The expert, the brief, the
10:16 simplifier. The point here is that you
10:18 are in control of your output length and
10:20 it is important that you include that in
10:24 the prompt. Eight on 10 providing a
10:26 scaffold. The rule here is you want to
10:28 give the AI you want to give Claude a
10:31 template or example to guide its
10:34 structure and style. So
10:36 let's the vague request would be
10:39 something like summarize this article.
10:41 But you know this is way better. Look at
10:43 this in the architected brief section.
10:45 Summarize the following article using
10:48 this format. Main thesis, one sentence.
10:50 AI fills this in. Key supporting points,
10:52 three bullet points. The AI fills this
10:55 in. And concluding insight, AI fills
10:58 this in. And then paste the text here.
11:00 So now you have this rigid structure
11:01 ensuring the summary is not only
11:04 accurate but is also formatted exactly
11:09 as needed. Um, so you know, really really
11:10 really
11:12 we want the best results. We want the
11:15 best final results and just by by saying
11:18 this the structure and style it helps.
11:20 It's a little bit of scaffolding goes a
11:22 long way.
11:26 Nine on 10 uh speaking the language. So
11:28 using advanced prompting terms can
11:30 trigger more sophisticated modes of
11:33 operation. So models
11:36 um are trained on a vast amount of text
11:38 about AI itself. So using terms from the
11:40 field activate specific powerful
11:42 behaviors. This is like cheat codes. So,
11:44 let's talk about a few power phrases
11:48 that Anthropic uh has literally told us
11:50 to use that most of us are not even
11:53 using. Think step by step. So, what's
11:55 the use case? Force the model to lay out
11:58 the reasoning process often leading to
12:00 more accurate results on complex
12:01 problem. So, you're going to want to
12:04 remember that one. Critique your own
12:06 response use case. Ask the model to
12:09 perform self-correction. Find flaws.
12:11 it's in in its initial draft and improve
12:15 it. Uh adopt the persona of an expert in
12:17 field use case primes the model to
12:19 response with a deeper more
12:22 domainspecific vocabulary and framework.
12:24 These are magic words. You might want to
12:27 screenshot this. Um this will help you
12:29 get the most out of it. And finally we
12:33 have uh the divide number 10. The divide
12:35 and conquer strategy. So the rule here
12:38 is for a complex task act as the con
12:40 conductor. Prompt for each part
12:43 separately then prompt for the
12:45 synthesis. So don't ask for a 10-page
12:47 report in a single prompt. Break it down
12:50 into logical subtasks that you manage
12:54 step by step. So if you listen to this
12:56 startup Ideas podcast, you know I love
12:57 ideas. You know I have a business
13:01 ideabous.com that gives away ideas. Um
13:04 and uh you know so you know I like
13:06 building businesses and you can see that
13:10 it it's easy to you know throw in hey
13:12 build a business plan as the one prompt
13:13 but this is a way better to do it. Step
13:16 one the blueprint create a detailed t
13:18 table of contents for a business plan
13:21 for a new specialty coffee shop. Step
13:24 two section by section write the
13:27 executive summary based on our plan. Now
13:29 write the market analysis section and
13:31 then the A dotted line indicates more
13:33 steps. And step three, the synthesis.
13:34 Review the complete business plan.
13:36 Ensure consistent tone and check for any
13:39 contradictions. So this is the project
13:41 workflow. You know, doing it section by
13:43 section, synthesizing, having the
13:45 blueprint, you know, you're dividing and
13:47 conquering. You're you're breaking out
13:49 to logical subtask. What might even be
13:51 helpful here is like literally taking
13:53 out a pen of paper and and being like,
13:55 "Okay, if I want a business plan, what
13:56 are the 10 things I need?" If I want to
13:58 create a deck, a fundraising deck, what
14:00 are the 10 things I need? If I want to
14:02 create a product that's, you know,
14:04 software product, what are the 10
14:06 different, you know, features I need?
14:07 This is going to be helpful for you to
14:10 get the most out of it. So, there you
14:13 have it. Uh, I've given you the 10 rules
14:15 of how you can get how you can prompt
14:18 the prompt claude to get literally 10x
14:21 more of it. This is the simple tools and
14:23 techniques that you can use that have
14:25 literally been shared by the people from
14:27 Anthropic themselves. We talked about
14:29 the tone of collaboration. We talked
14:31 about uh which is using friendly, clear
14:33 and firm tone. We talked about the
14:34 principle of explicitness. We talked
14:36 about the defining the boundaries. We
14:37 talked about the exploratory draft. We
14:39 talked about specifying the details. We
14:41 talked about explaining the why. We
14:43 talked about how important brevity is,
14:45 why you can put a should put a little
14:47 scrap scaffolding, how how important it
14:50 is to speak the language. And we talked
14:51 about the divide and conquer strategy.
14:53 So what does that look like when you put
14:56 it all together? This is, you know, when
14:59 you came into this episode, you might
15:01 have prompted or maybe not, but you
15:04 might have prompted Claude saying
15:05 something like, "Tell me about
15:07 stoicism." By the way, I call it Clode,
15:09 but for you, I'll I'll call Claude cuz
15:12 people get mangry and mad at the an uh
15:15 in the comment section about that.
15:16 Let's talk about now now that you know
15:19 the tools of how to prompt Claude
15:21 better, let's let's let's see it at
15:23 work. Instead of saying, "Tell me about
15:25 stoicism," you're going to say, "Act as
15:27 a university professor of philosophy."
15:29 Why? Because you've included the persona
15:30 in there. You're going to say, "I'm
15:32 preparing a 1-hour intro lecture for
15:34 students with no prior knowledge."
15:36 That's going to explain the why. You're
15:38 going to say, first create a lecture
15:40 outline with three main sections, the
15:42 divide and conquer. The outline should
15:44 have a clear introduction and body.
15:47 You've now constrained the scope and
15:49 conclusion. Please format this as a
15:54 nested bulleted list rich output. For
15:56 each major point, include a key stoic figure,
15:58 figure,
16:00 example, senica,
16:02 uh, explicitness,
16:05 um, and one of their core ideas. Your
16:08 tone should be accessible and engaging
16:10 tone. So, you've included that all in
16:13 there. And then the output is just way
16:16 better. So, these are little things to
16:19 help you get the most out of uh, these
16:21 LLMs. And like why it matters is because
16:23 we're all trying to get the most out of
16:25 it. Uh we're all trying to create
16:28 products that people love. We're all
16:29 trying to create content that people
16:31 love. We're all trying to do things that
16:34 people love. And a lot of the times we
16:36 get AI slop. And this is going to help
16:39 you not get AI slop. And it's and that's
16:42 why I put the time to scour the internet
16:45 for you so you can get the most out of
16:48 this pro uh product. So so powerful. I
16:51 have no affiliation with Anthropic. Um,
16:54 but I want I want to see you not just
16:56 like, comment, and subscribe on this,
16:59 but actually go and build something
17:02 great with this stuff. So, um, I'm
17:04 rooting for you. Have a creative day and
