Latest / AI Fire Daily / #91 Robin: The "Write First, Design Second" Rule - How to Build Killer Decks with Claude and Replit Design

Transcript

#91 Robin: The "Write First, Design Second" Rule - How to Build Killer Decks with Claude and Replit Design · AI Fire Daily

• 0:00
  We have, we've all been there. You type a prompt
• 0:03
  into an AI slide generator. Beat 30 seconds later,
• 0:07
  it spits out this gorgeous color coordinated
• 0:10
  deck. Oh yeah, and it always looks so perfect
• 0:12
  at first glance. Right, you think, like, amazing,
• 0:15
  I am done. But then you actually read. the slides,
• 0:19
  and you realize the underlying story is a total
• 0:21
  mess. The logic is just completely broken. So
• 0:24
  what happens, you spend the next three hours
• 0:26
  manually dragging text boxes around trying to
• 0:29
  salvage the narrative. It's incredibly frustrating.
• 0:31
  You get these beautiful templates and perfect
• 0:33
  gradients. The visual polish is definitely there,
• 0:36
  but the actual core message. it's just entirely
• 0:41
  weak. Welcome to this deep dive. Today we're
• 0:44
  looking at a dramatically better workflow. We
• 0:46
  are unpacking a surprisingly simple approach
• 0:49
  that completely solves this exact AI presentation
• 0:52
  trap. The secret is that you just separate the
• 0:55
  writing from the design. Just a quick reminder
• 0:58
  for you listening, this is a deep dive, not a
• 1:01
  standard interview. We're going to meticulously
• 1:03
  break down this workflow step by step so you
• 1:06
  can actually use it. Right. we're looking at
• 1:08
  a complete inversion of how most people instinctively
• 1:12
  use these tools. Like, people usually rush straight
• 1:15
  into the visual layout because it's fun and immediate.
• 1:18
  But if your underlying argument is weak... There
• 1:21
  is no visual template on earth that can save
• 1:24
  it. Exactly. Beat. So let's start by unhacking
• 1:28
  the foundational rule here. Write first, design
• 1:31
  second. The guide we're studying is very specific
• 1:33
  about this mechanism. You write the entire presentation
• 1:36
  in Claude first. Only then do you design it in
• 1:39
  Replet Design. It sounds so simple when you say
• 1:41
  it out loud, but it fundamentally changes the
• 1:44
  entire output. You're letting Claude handle the
• 1:46
  logic. Then you're letting Replet handle the
• 1:49
  layout. By doing this, you're forcing the AI
• 1:52
  to plan the whole presentation as pure text long
• 1:55
  before any visuals even exist to distract it.
• 1:58
  And it all starts with what the source material
• 2:00
  calls a master prompt. Let's define what that
• 2:02
  actually means in practice. A master prompt is
• 2:05
  basically a reusable template where you only
• 2:07
  change the blanks. Exactly. You don't need to
• 2:10
  reinvent the wheel and write a brand new prompt
• 2:11
  every single time you need a deck. Great. You
• 2:13
  just keep the core structural instructions. Then...
• 2:16
  You fill in five specific fields for your new
• 2:19
  topic. Those fields are role, topic, slide count,
• 2:23
  audience, and tone. I really want to focus on
• 2:26
  that first one for a moment, the role. Oh, it's
• 2:28
  so important. The guide says this is where you
• 2:30
  should actually spend the most time thinking.
• 2:32
  Beat and honestly, I completely agree with that.
• 2:35
  I still wrestle with prompt drift myself, you
• 2:38
  know when I don't set a strict role Oh, absolutely
• 2:40
  prompt drift is just when the AI forgets its
• 2:42
  original instructions and Wanders off -topic.
• 2:45
  It happens so easily. Yeah setting a very specific
• 2:48
  role anchors the AI Yeah, it gives the language
• 2:51
  model a rigid persona to operate within right
• 2:54
  The text gives a fantastic example here. Think
• 2:57
  about the difference between asking Claude to
• 2:59
  act as a market analyst versus acting as a startup
• 3:02
  founder. They are completely different lenses
• 3:04
  for looking at the exact same data. Exactly.
• 3:07
  If you tell Claude to act as a market analyst,
• 3:10
  its whole operational mechanism shifts. It focuses
• 3:14
  on historical trends. It looks at the raw numbers.
• 3:18
  It provides sober objective evidence. The language
• 3:21
  is going to be very measured and precise. Right.
• 3:23
  But if you say, act as a startup founder pitching
• 3:26
  to venture capitalists, The focus shifts entirely.
• 3:30
  It suddenly becomes all about the vision. Yes.
• 3:32
  It pushes urgency. It explains exactly why people
• 3:35
  should care right now. It's the exact same topic.
• 3:38
  But because of that role, you get a totally different
• 3:40
  structural narrative. And that leads right into
• 3:42
  the audience and tone fields. Your audience changes
• 3:46
  everything about how an argument needs to be
• 3:48
  structured. Totally. Like an investor deck requires
• 3:50
  market proof and immediate business impact. Right.
• 3:53
  But a deck for university students. That needs
• 3:56
  to slow down the logic. Right. It needs simpler
• 3:59
  analogies and more foundational examples. Jumping
• 4:01
  straight into a slide designer without figuring
• 4:03
  all of that out first. It's like painting a house
• 4:06
  before the framing is even up. That is a perfect
• 4:08
  analogy. The underlying structure just isn't
• 4:11
  there to support the paint. Beat. So let me ask
• 4:14
  you this. Why force text first? Why not just
• 4:18
  jump in and edit the text right there inside
• 4:20
  the visual generator? Because visual distraction
• 4:23
  ruins logical flow. It's just human psychology.
• 4:27
  When you're looking at font pairings and color
• 4:29
  palettes, your brain literally stops thinking
• 4:32
  about the strength of the argument. You start
• 4:35
  worrying about how a sentence fits inside a little
• 4:37
  blue rectangle. Forcing it into Claude first
• 4:40
  makes you build a clear logical reason for why
• 4:43
  each slide belongs. It forces a strong argument
• 4:46
  before shiny design distracts us. Precisely.
• 4:49
  So let's explore how we actually sculpt this
• 4:51
  argument inside Claude. Let's do it. The guide
• 4:54
  mentions choosing the slide count on purpose,
• 4:56
  which I found fascinating. It says your slide
• 4:58
  count isn't just a random number you pick. It
• 5:00
  actively decides how much heavy lifting each
• 5:03
  slide has to do. This is a really mathematical
• 5:05
  way to limit cognitive load. Cognitive load is
• 5:08
  just the amount of mental effort needed to understand
• 5:11
  new information. Think about how an AI model
• 5:14
  generates text. If you ask Claude for 20 slides
• 5:18
  on a simple topic, it's going to spread the information
• 5:20
  too thin. Each slide makes a tiny incremental
• 5:24
  point. Right, the whole argument gets diluted.
• 5:27
  Exactly. But if you restrict it and ask for exactly
• 5:30
  six slides, Quad is computationally forced to
• 5:33
  cut the weak ideas and only keep the absolute
• 5:36
  strongest points. The source recommends around
• 5:38
  10 slides for a normal standard presentation.
• 5:42
  They say 10 slides is usually the perfect amount
• 5:44
  for about 15 minutes of actual talking. Now I
• 5:48
  have to push back on that 10 slide rule just
• 5:50
  a little bit. Is 10 slides too rigid? What if
• 5:53
  I'm pitching a complex, highly technical B2B
• 5:57
  sauce product? 10 slides feels dangerously thin
• 6:00
  for that. Am I just supposed to leave out all
• 6:02
  the vital technical specs? It's a fair question,
• 6:06
  but you aren't leaving them out. You're just
• 6:07
  shifting where they live. You have to remember
• 6:09
  that the slide deck is just the narrative spine
• 6:13
  of your presentation. It is not the entire encyclopedia
• 6:16
  of your product. If you need a deep dive into
• 6:19
  technical architecture, that belongs in a separate
• 6:21
  white paper or a technical appendix. Right. That
• 6:24
  makes sense. 10 Slides forces a complete persuasive
• 6:28
  argument without making the deck feel endless.
• 6:31
  It gives Claude enough room to build the core
• 6:33
  logic, and later, it gives Reclit Design enough
• 6:37
  physical space to make each slide feel clean
• 6:39
  and readable. Okay, I see. You're forcing narrative
• 6:42
  discipline on the AI. Exactly. Now, this next
• 6:45
  detail might honestly be my favorite part of
• 6:47
  the entire workflow. It's the instruction to
• 6:50
  tell Claude to write all slide titles as key
• 6:52
  takeaways. Yes. This is a massive structural
• 6:55
  upgrade to how most people build presentations.
• 6:58
  Totally. Think about a standard bad slide title.
• 7:01
  It's usually just a lazy label. Something like
• 7:03
  market size or Q3 growth. Right. it tells the
• 7:06
  audience absolutely nothing about what the actual
• 7:08
  data means. Right. If the slide just says market
• 7:11
  size, the audience spends the next 45 seconds
• 7:14
  squinting at a pie chart trying to figure out
• 7:17
  what the point is. Yeah, completely ignoring
• 7:19
  what you're actually saying out loud. Exactly.
• 7:21
  But a better slide title is a full, complete
• 7:24
  thought. Something like... The creator economy
• 7:27
  has doubled to 250 billion in two years. The
• 7:31
  title itself carries the entire message. Exactly.
• 7:34
  You want to explicitly instruct Claude in your
• 7:37
  master prompt. Write every single slide title
• 7:40
  as the main takeaway, not as a generic label.
• 7:43
  It instantly stops the deck from looking like
• 7:45
  a random disconnected list of topics. It forces
• 7:48
  the AI to make an assertion on every single page.
• 7:51
  So, why are takeaway titles so critical to this
• 7:54
  whole process? Because of natural human skimming
• 7:57
  behavior. We know people simply do not read every
• 7:59
  single word on a slide. They scan. Right. They
• 8:02
  look at the biggest text first. If your title
• 8:04
  is the main point, they get the crucial message
• 8:06
  instantly. Even if they ignore your bullet points
• 8:08
  entirely, they still understand your argument.
• 8:11
  Skimmers still grasp your entire narrative without
• 8:14
  reading the bullet points. Exactly. And that
• 8:16
  ties perfectly into the next crucial instruction
• 8:18
  for Claude. Okay. actively force it to build
• 8:22
  one continuous argument. Right. The guide says
• 8:25
  a good presentation shouldn't just be isolated
• 8:27
  facts. It should move in one deliberate direction.
• 8:32
  You start with a pressing problem. You show what
• 8:34
  is changing in the world. You add undeniable
• 8:37
  proof. You explain why it matters right now.
• 8:40
  And you end with a clear action. To get that,
• 8:43
  you literally add this exact line to your prompt.
• 8:45
  Build the presentation as one continuous argument.
• 8:49
  Each slide should directly set up the next slide.
• 8:52
  Without that line, a language model will often
• 8:54
  just give you 10 perfectly true but totally disconnected
• 8:58
  bullet points about a topic. And we absolutely
• 9:00
  cannot forget speaker notes. You have to explicitly
• 9:03
  ask Claude to generate the actual words you will
• 9:05
  say out loud. Oh, for sure. This is the secret
• 9:07
  to keeping the slides from becoming giant, unreadable
• 9:10
  walls of text. It is the final piece of the text
• 9:13
  puzzle. So now you have a slide title, a main
• 9:16
  message, a rough layout idea, and detailed speaker
• 9:19
  notes. At this point, you sit down and read the
• 9:22
  full draft in Claude. You cut any repetitive
• 9:25
  slides. You fix the flow of the argument right
• 9:28
  now in pure text while it only takes seconds
• 9:31
  to change. Because fixing a broken argument after
• 9:34
  the slides are already beautifully designed is
• 9:37
  an absolute nightmare. It really is. So once
• 9:39
  the text is perfect, you ask Claude to generate
• 9:42
  a final ready -to -paste prompt. You take that
• 9:45
  massive block of text and you are finally ready
• 9:47
  to open Replet Design. Sponsor. OK, we are back.
• 9:52
  We have our text blueprint perfectly sculpted
• 9:54
  and refined and clawed. Now we cross the bridge
• 9:56
  into the visual realm. We let the AI actually
• 9:59
  do the heavy lifting of design. This is where
• 10:02
  Replet Design really takes over the process.
• 10:04
  The hard structural thinking is already done.
• 10:07
  Right. Replet already knows the exact slide count.
• 10:09
  It knows the audience, the tone, the narrative
• 10:11
  structure, and the core message. You just open
• 10:14
  up Replet Design, you choose the Slides option,
• 10:16
  you paste that massive pre -formatted prompt
• 10:18
  from Claude, and you hit Send. That's it. Beat.
• 10:23
  The guide says the very first visual version
• 10:24
  is usually built in about 30 seconds. 30 seconds.
• 10:27
  It is incredibly fast. You watch it happen. Yeah.
• 10:31
  The chat interface on the left shows the AI actively
• 10:34
  planning the components. The canvas on the right
• 10:37
  shows the actual slides building in real time.
• 10:41
  It's wild. The charts render cleanly. The numbers
• 10:43
  are perfectly readable. It's doing the work of
• 10:45
  a junior designer instantly. Whoa. Imagine generating
• 10:48
  an entirely new visual logic in 30 seconds. Two
• 10:52
  secs silence. Mm -hmm. It really is like having
• 10:55
  a junior designer sitting right next to you just
• 10:56
  waiting for instructions. It really is. But the
• 10:59
  crucial part of this workflow is understanding
• 11:01
  what you do next. Right. That first 30 second
• 11:03
  version. It will almost certainly not be perfect.
• 11:06
  Sure. It is just a starting direction. And the
• 11:09
  guide has a golden rule for this exact moment.
• 11:11
  Yes, and it's a huge mindset shift. If the overall
• 11:14
  visual direction is completely wrong, do not
• 11:17
  try to fix every slide manually. Snow. Just rebuild
• 11:20
  the prompt. A complete rebuild takes another
• 11:22
  30 seconds. Yay. Manually dragging text boxes
• 11:25
  around to fix a bad layout takes all afternoon.
• 11:28
  A strong, detailed Claude prompt gives Replet
• 11:32
  Design far fewer chances to guess wrong in the
• 11:35
  first place. Right. But if it does misinterpret
• 11:38
  the vibe, you just adjust the core instruction
• 11:40
  and generate it again. Now, once you have a layout
• 11:43
  you generally like, you can start tweaking. There
• 11:45
  is a simple edit button for tiny little word
• 11:48
  swaps, but the text chat feature is where the
• 11:51
  real magic happens. You can click on a specific
• 11:54
  element on the slide, and you literally tell
• 11:56
  the AI what to fix in plain English. But there
• 11:59
  is a catch. You cannot be vague. If you just
• 12:02
  type, make this better, you are giving the AI
• 12:05
  agent way too much room to guess what you mean.
• 12:07
  it might change the font when all you wanted
• 12:09
  was a different background color. Right. You
• 12:11
  have to be specific with your commands. You say
• 12:13
  something like, make the main percentage number
• 12:15
  the very first thing people notice. Exactly.
• 12:17
  Or the text block is way too close to the left
• 12:20
  edge. Add more padding and spacing on the left
• 12:23
  side. Exactly. You name the actual problem. The
• 12:26
  AI goes into the code and changes that exact
• 12:30
  spatial element. It does not blindly redesign
• 12:33
  the whole slide and ruin the parts you already
• 12:35
  liked. And what's wild is that you can ask for
• 12:37
  much bigger structural changes, too. Oh, yeah.
• 12:39
  Let's say a slide is just a dense wall of statistics
• 12:43
  in paragraph form. You can just ask Replet to
• 12:46
  turn that text into a simple three column bar
• 12:49
  chart. Or if a single slide feels way too crowded
• 12:51
  with information, you just ask the AI to split
• 12:54
  the content perfectly across two slides. Right.
• 12:57
  If the visual hierarchy feels flat and boring,
• 13:00
  you ask it to make the key data point three times
• 13:03
  larger and bold. I want to pause here because
• 13:05
  this is a big distinction. Why use the text chat
• 13:08
  to fix a visual layout instead of just clicking
• 13:11
  and dragging the elements yourself? Because the
• 13:13
  chat edits the underlying structural code of
• 13:15
  the presentation. Okay. Clicking and dragging
• 13:17
  is just applying a manual, superficial style.
• 13:20
  It's the difference between applying a band -aid
• 13:22
  to a symptom versus curing the underlying disease.
• 13:26
  When you ask the AI to turn a paragraph into
• 13:28
  a chart, it fundamentally rewrites the DNA of
• 13:32
  the slide. You simply cannot do that by just
• 13:34
  dragging a text box across the screen. Chat fixes
• 13:37
  the core structure, saving you hours of tedious
• 13:40
  manual tweaking. Right. And once that structure
• 13:42
  finally looks perfect, you save the deck as a
• 13:45
  custom template. You start building a small personal
• 13:48
  library of reliable layouts. If you add your
• 13:51
  specific brand fonts and brand colors just one
• 13:54
  time, All your future decks will follow your
• 13:57
  brain guidelines automatically. That is massive
• 13:59
  for maintaining consistency across a team. Oh,
• 14:02
  totally. So the deck is written. It is beautifully
• 14:04
  designed. It is cleaned up. Now we finally reach
• 14:07
  the last mile of the process. Sharing it? Yes.
• 14:11
  Beat. Because a perfect, brilliant deck is completely
• 14:14
  useless if it just lives in a silo on your desktop.
• 14:17
  Ripley Design actually offers four very simple
• 14:20
  sharing options. Present. Export. collaborate,
• 14:25
  and publish. Presenting is exactly what it sounds
• 14:27
  like. It's full screen mode for when you're in
• 14:29
  the room. Exporting is downloading the deck as
• 14:31
  a standard PowerPoint file or a PDF. Collaborate
• 14:34
  lets your teammates jump in and leave comments.
• 14:37
  But I want to really dank dive into that fourth
• 14:40
  option, the publish feature. This is arguably
• 14:43
  the most powerful functional option in the entire
• 14:45
  tool. Hitting publish generates a live hosted
• 14:50
  web link for your presentation. You can open
• 14:52
  that link from any device in the world. Even
• 14:54
  on your phone's mobile browser, it renders perfectly.
• 14:57
  And here is the kicker. When you change a slide
• 15:00
  on your end, that exact same link updates instantly
• 15:04
  for everyone else. Amazing. The psychological
• 15:07
  relief of this feature is honestly huge. Think
• 15:10
  about finding a massive typo on slide four just
• 15:13
  five minutes before a critical client meeting.
• 15:15
  Oh, the worst feeling. Normally, that is a full
• 15:18
  -blown crisis. You have to fix it, re -export
• 15:20
  the PDF, draft a new email, and pray the client
• 15:22
  opens the new attachment instead of the old one.
• 15:25
  But with a live link, you just quietly fix the
• 15:28
  typo and repel it, and the client screen updates
• 15:31
  immediately. It completely eliminates the nightmare
• 15:34
  of outdated files. No more messy, confusing email
• 15:38
  threads. No more attachments, frantically named,
• 15:42
  finalv4definitelyfinal .pdf floating around the
• 15:45
  office network. What makes this Live Link fundamentally
• 15:49
  different from just sharing a Cloud Drive folder
• 15:52
  with your team? Well, a Cloud Drive folder still
• 15:54
  relies on static independent files. You are still
• 15:58
  opening file, saving a file, syncing it. Right,
• 16:01
  and hoping everyone clicked the most recent timestamp.
• 16:04
  Exactly. A Live Link is completely different.
• 16:06
  It acts as a single centralized source of truth
• 16:09
  that renders dynamically in the web browser in
• 16:11
  real time. It permanently kills outdated files
• 16:14
  and messy email threads for your whole team.
• 16:16
  Exactly. It keeps everyone everywhere perfectly
• 16:18
  synced without any extra effort. Let's zoom out
• 16:20
  for a second. Let's synthesize this entire journey
• 16:22
  we just went on. The big takeaway here is the
• 16:25
  total inversion of the normal workflow. Right.
• 16:27
  And the strict discipline it actually takes to
• 16:29
  maintain that inversion. Right first. Design
• 16:31
  second. AI is a logic engine first. And a design
• 16:34
  engine second. You separate the story from the
• 16:37
  visual. Build the argument. Then build the deck.
• 16:40
  It completely shifts how we interact with these
• 16:42
  tools. It really does. I want to encourage you,
• 16:45
  listening right now, to actually try this workflow.
• 16:48
  Apply this master prompt to framework to whatever
• 16:50
  big idea or pitch you are wrestling with this
• 16:53
  week. Force the logic first in Claude. See how
• 16:57
  much time it actually saves you when you aren't
• 16:59
  fighting with templates. Beat. But going through
• 17:02
  this process also leaves me with a much bigger,
• 17:05
  slightly more philosophical question. Something
• 17:07
  for you to mull over as we wrap up. If AI can
• 17:11
  now perfectly structure our complex arguments.
• 17:14
  and beautifully design our visual slides in seconds?
• 17:17
  Doesn't that shift all the pressure right back
• 17:19
  onto our human presence? That's a fascinating
• 17:21
  point to consider. The slide deck is no longer
• 17:23
  a crush you can hide behind. Exactly. If the
• 17:26
  deck is flawless every single time, how does
• 17:29
  this change the way we actually speak and perform
• 17:31
  in the room? The presentation is fixed. The logic
• 17:34
  is crystal clean. You don't have to waste your
• 17:36
  energy manually fixing those text boxes the night
• 17:38
  before. But now you actually have to stand up
• 17:41
  and deliver. The spotlight is entirely back on
• 17:43
  you. It really is. We will leave you with that
• 17:46
  thought. Thank you for diving deep with us today.
• 17:49
  Out T -Row music.