Strategy Automation: Build Your Own AI App with PartyRock by AWS
Introduction
Building AI Apps that deliver immense value doesn’t have to be hard. In fact, it can be done in 30 minutes! I’ve shared several methods for developing AI apps on my blog, but this one is probably the most accessible for non-techies. I hope it helps you think differently about building your own tools with AI.
—
Hey readers, today we are making a simple yet useful AI app with PartyRock.
This app is an ideation assistant that specializes in creating thought starters for music/brand partnerships.
All we need to do is give it an artist, a brand, and a little brief with some partnership details.
From there it will produce a robust set of ideas that you can then take away to workshop on your own or with your team.
Check out the finished AI App here and read on to make your own.
Here's what we'll cover together:
The thought process that comes before we start building an AI App
How to engineer prompt chains for complex / multi-step AI workflows
Creating an AI App with PartyRock
Let's jump in.
Can AI help me plan brand partnerships?
When done correctly, brand partnership campaigns are a source of fresh and evocative storytelling that fans go crazy for.
These campaigns are tailor-made to capture our attention, bridging unlikely combos such as Post Malone / Pokemon or Lil Nas X / League of Legends.
However, the inherent challenge in partnership work is that it's highly contextual to the two (or more) entities that we're combining.
It has to feel right, and getting to that point takes a lot of creative critical thinking.
So how do we fit together the "jagged edges" of each unique brand identity in a partnership so the whole is greater than the sum of its parts?
Well, AI can help us with that.
The thought process for AI app design
Whether you're new to AI or already experienced, many still struggle to effectively use AI & LLMs (Large Language Models) in their day-to-day work.
This is because many of us don't realize how abstract our "work" is to a machine.
Thinking like AI vs. thinking like a human
Let's examine the task of preparing thought starters, an essential step in developing a partnership campaign.
Off the top of your head: do you know how many discrete steps or tasks we complete when preparing thought starters?
Probably not. But you still know how to get it done:
Do a little research
Jot down some ideas
Jump into Miro, mind map it a bit
Produce some draft ideas that your team can discuss
For language models like ChatGPT or Claude, it's not so easy.
And this is also one reason why building a context map is so helpful when working on more complex tasks.
Be specific: how to leverage AI as an effective helper
When you give an LLM a complex task that requires multiple steps, it doesn't have an inner monologue that says, "Great, I've completed the first step. What is the smartest next step I should tackle next?"
Instead we need to offer richer context & specificity around your request.
Always break complex tasks down into a list of smaller, discrete tasks that can be implemented in sequence.
By atomizing the big picture tasks into more bite-sized pieces, we are able to get higher quality outputs much more consistently.
Why can't AI just figure out what I want when I ask?
This is due in part to the nature of how GPTs (Generative pre-trained transformers) work. They rely heavily on next-token prediction, which is a stochastic or non-deterministic process.
For example, let's unpack this Q&A scenario wherein we're sending a text completion request to our language model AI assistant.
[User]: Tell me a joke.[AI Assistant]: Why did the... The format of this request will have our assistant thinking about how to complete the joke.
The way it picks the next word is through probability with a little bit of randomness. The assistant may think there's a high likelihood, say 90%, that the next word should be chicken, because that's what it can infer from all the chicken jokes in its training data.
But there's also a very low chance that the next word is bagpiper, maybe 0.1%.
If you ask for a joke 1000 times with this set up, it might produce "why did the chicken..." 900 times, but only once would it joke about a bagpiper.
If your prompt was "Tell me a joke about Scotland", the odds of bagpiper being the next word instead of chicken will go up because of the context you're giving to the assistant with your request.
“This nuance is why it’s important to have a tight focus for the function that an AI app will perform. In our case, the focus is on one subtask of a larger task: ideating partnership concepts.”
AI Apps: Different than AI Chatbot Interfaces
Rather than perform this task in a chatbot interface, we are using an AI App to submit a sequence of prompts via API where each prompt contains its own content and context.
In this way we have more control over each step of call & response between us and the AI.
We can reference previous answers, inject new context or considerations, and more.
We can't do this as easily in a chatbot style interaction because the AI model retains context from the conversation, so context-switching comes with a bit of friction and reframing.
You have to say things like "Ok, I no longer want you to act as a senior marketing strategist anymore. You're a Gen Z athlete! Think like you're Gen Z!"
^While this is a bit of a hyperbole, it holds true that each new chat with a language model is a blank slate in terms of context.
With an AI app, we can take advantage of the blank slate we get with each message we send to ensure that every step in our prompt chain is specific to the task and the mindset required for completing it at a high level.
When should I use AI apps vs chatbots?
If you have a multi-step task that is a regular part of your personal or professional workload, it makes sense to create an AI app instead of firing up a chat every time you want to work through it.
Here's why.
By putting the discrete steps of your process on autopilot with a sequence of specialized prompts, you can quickly complete the task with AI while still getting a highly consistent quality of response.
In contrast, if you were to use a chatbot and work through this process off the cuff each time you need it done, it's much harder to analyze how you can improve your AI workflow.
Plus you are reinventing the wheel every time you do the task, instead of logging the best prompts for each portion of the task and stringing them together so that they can be automatically executed every time you pass some context into the AI app.
Prompt Engineering: creating prompt chains for complex tasks
As we outlined at the top of this article, we are building a partnership ideation app to help us come up with smart, creative thought starters that work for both the artist and the brand.
For humans, this is a whole brain task that comes naturally to us. We creatively think, we analyze, we improve the ideas, voila.
But for AI, each of these tasks is a discrete cognitive function.
Prompt Chain Essentials: Ideate, Analyze, Refine
To get a good response, we will separate the task of thought starter creation into three discrete cycles of thought:
Ideation: Come up with a first batch of ideas based on the user's criteria that creatively unpacks the problem space in a variety of ways
Critical Analysis & Recommendation: Critique the ideas from multiple perspectives, think laterally, and suggest improvements to the ideas based on the critique
Refinement: Produce a revised set of ideas that effectively incorporate the critique and recommendation from step 2.
If we were to write all of this into a single prompt it might look like this hot mess:
"Here's the context: {INFO}. Now come up with a set of ideas. Critique them and suggest improvements. Then produce a refined set of ideas based on your recommendation."
But it's more effective to build a chain of prompts that automatically hits each discrete step in the thought process, like so:
[Step 1] – "Based on the following context, draft three thought starter concepts for a brand partnership campaign between ARTIST and BRAND. {PASTE ARTIST / BRAND DETAILS HERE}"
[Step 2] – "Critically analyze this set of thought starters within the context of a brand partnership between ARTIST and BRAND. Provide a critique for each concept. Offer insights from the perspective of both a superfan and a devil's advocate on how to improve the thought starters. {STEP 1 CONTEXT}"
[Step 3] – "Here is a set of ideas for a brand partnership between ARTIST and BRAND: {STEP 1 CONTEXT}. My team reviewed these ideas and this is their feedback: {STEP 2 CONTEXT}. Synthesize the ideas and feedback into an improved and updated set of ideas.
These prompts are simplified, but you get the gist.
Subtasks of subtasks
As you can see, even within these atomized steps my prompts include multiple components.
In Step 2 we have criteria such as:
Analyze these ideas
Critique each concept.
Offer insights from multiple perspectives
With a tool like PartyRock, you can get as granular as you'd like with atomizing the steps.
At a certain point, though, you do reach a stage where it's superfluous to continue breaking subtasks into smaller subtasks.
With that being said, the biggest benefit to this atomic step-based prompting with AI is that each request & response is a standalone nugget of context that you can inject into other prompts.
While our AI app is linear, flowing from Step 1 to Step 3 in a straight line, we could instead design it to operate in multiple parallel paths.
For example it could take the context we give it and simultaneously generate ideas that work for the artist (Step 1A), the brand (Step 1B), while also doing research into the top performing brand partnerships from 2023 (Step 1C).
All of this would be happening in parallel, triggered by the original context.
Then in Step 2 we recombine 1A/1B/1C into a synthesis step ahead of moving to a critique stage.
You can see how this gets complex quickly, but how it also starts to more accurately reflect the way that we humans critically think about these challenges
Designing our AI App in PartyRock
Let's start building our 3 step AI App in PartyRock.
To start, make an account via PartyRock.aws. Once that's done, click build your own app on the PartyRock homepage.
In the popup, click Start from an empty app.
You'll land on a screen like the one below. Mouse over the dotted area and click where it says Create Widget.
Creating the User Input widgets
We'll start by adding three User Input widgets.
For each widget, add details via the sidebar that appears.
Title your widget, but you don't need a placeholder or default value for our app.
Here is how I arranged my three input widgets.
Creating the AI Text Generation widgets
Now we are going to add three text generation widgets.
For each widget, add details via the sidebar that appears.
For the first step, I've titled it Generate Thought Starters and I'm using Claude as the AI model. You'll also see in the prompt area that I've linked the User Input widgets using the "@" symbol. This is how we can inject the user's context into the prompt so that our AI generated text is relevant to our request.
Here are the raw prompts I've used in each AI Widget step, which have been simplified from their normal level of detail for the purposes of this article.
AI Widget Step 1, Generate Thought Starters
Based on the following context, draft three thought starter concepts for a brand partnership campaign between the artist and the brand I've highlighted in the context below. Aim to produce a robust and lengthy output for each concept, atomizing the details of each idea as they pertain to the overarching partnership context.
Artist: [Artist Name]
Brand: [Brand Name]
Partnership Context: [Context Widget]AI Widget Step 2, Critique and Refine Thought Starters
Critically analyze this set of thought starters within the context of a brand partnership between [Artist Name] and [Brand Name].
Produce a richly detailed and nuanced critique that digs into how these ideas map back to the key areas outlined in the partnership context. Your response should include a holistic critique of the entire suite of concepts, as well as an individual critique of each concept.
As part of your response, analyze the concepts from the perspective of both a superfan and a devil's advocate to enrich your critique.
Thought Starters: [Step 2: Generate Thought Starters]
Partnership Context: [Add your partnership context here]AI Widget Step 3, Generate Improved Thought Starters
You are being given a first draft set of thought starters for a brand partnership between [Artist Name] and [Brand Name]. You are also being given a well-informed critique / analysis of those thought starters.
Your task is to produce a well-organized Word document that syncretizes the original ideas and feedback into an improved set of concepts.
Synthesize a robust and lengthy output for each concept. Atomize the details of each idea to further elucidate its role in the overarching partnership context. In addition to your synthesis, include a section per concept that highlights strategic & tactical considerations.
Format:
- Include an executive summary
- Minimum acceptable response length is 500+ words
Readers of the final response should be able to understand the thought starter ideas without referencing any prior material.
First draft thought starters:
[Generate Thought Starters]
Critique and analysis of thought starters:
[Critique and Refine Thought Starters]
Brand partnership context:
[Add your partnership context here]Double-checking the final AI App build
Once you've created your 3 text generation widgets, you can arrange your app interface however you want.
Here is how mine is laid out.
You can see in the description text that the first AI widget is waiting for us to fill in the User Input fields before it will begin generating.
And the second and third AI widgets are waiting for the previous AI steps to complete before they begin running as well.
Running the AI App
Ok, I'm excited to run this AI App and see what we get.
First, let's gather the context we're going to feed into the app.
For this example I have selected Mac Demarco as our Artist and Apple / MacBook Pro as our brand.
I have also used ChatGPT to prepare a mock partnership brief that will serve as our partnership context.
Below is the brief that ChatGPT produced for me. You can easily swap this for your own campaign brief, or even use just a single sentence as your context. It doesn't need to be this robust to still produce great results.
Brand Partnership Brief: Mac DeMarco and Apple
Campaign Title: "Unleash Creativity: Mac DeMarco x Apple"
Objective: To showcase the synergy between Mac DeMarco's unique musical style and Apple's cutting-edge technology, emphasizing creativity and innovation.
Target Audience: Tech-savvy music enthusiasts, aged 18-35, who appreciate both innovative technology and indie music.
Campaign Overview:
Collaboration Theme: "Simplicity in Creation." The campaign will focus on how Apple's technology can simplify the creative process, enabling artists like Mac DeMarco to produce unique sounds effortlessly.
Key Messages: Apple technology is an enabler of artistic expression. Creativity knows no bounds with Apple’s intuitive design and features.
Key Deliverables:
Exclusive Track Creation: Mac DeMarco to produce a track using only Apple devices and software, showcasing the process in a mini-documentary.
Interactive Ad Series: Ads featuring behind-the-scenes footage of Mac DeMarco's creative process, focusing on the use of Apple products.
Social Media Takeover: Curated content by Mac DeMarco on Apple's social platforms, including live sessions, Q&As, and sneak peeks into his creative journey with Apple products.
In-Store Experience: Limited-time interactive installations in select Apple stores where fans can create music using the same setup as Mac DeMarco.
Merchandise Collaboration: Limited edition Mac DeMarco-themed Apple accessories.
Timeline:
Pre-Launch Teasers: Q3 2024
Campaign Launch: Q4 2024
In-Store Experience and Merchandise Release: Q4 2024 – Q1 2025
Budget: TBD
KPIs:
Engagement Rates (Social Media, Ad Views)
Sales Impact (Apple products used in the campaign)
Brand Sentiment Analysis
Earned Media Value
Legal and Compliance:
Rights for Music Usage
Brand Guidelines Adherence
Conclusion:
This partnership aims to bridge the gap between technology and creativity, showcasing how Apple's intuitive design can be the cornerstone of artistic innovation, as seen through the lens of Mac DeMarco's unique creative process.Alternately, your prompt could be as simple as "A brand partnership between ARTIST and BRAND focused on co-creation, artificial intelligence, and pepperoni pizza."
Test out different ways of structuring your partnership context to see how it alters what your AI app generates.
Step 1: Our first set of Thought Starters
Here is what Claude produced as the first step of our AI App, using the prompt and context we set for it.
Here are three thought starter concepts for a brand partnership campaign between Mac DeMarco and Apple/MacBook Pro: Concept 1: The Studio Session - Mac DeMarco records an entire album exclusively using MacBook Pro, GarageBand, and Apple peripherals. - Apple produces a documentary-style ad showcasing Mac's full creative process - writing, recording, producing, and mixing the album. Voiceover discusses how the MacBook Pro's processing power, GarageBand's features, and seamless ecosystem enables swift idea fruition. - Documentary aired as YouTube pre-roll ad and embedded on Apple Music. Shorter versions run as TV commercials. - Behind-the-scenes footage and album snippets teased on social media by both Mac and Apple ahead of release. - Album made available exclusively on Apple Music first, with Apple-exclusive deluxe vinyl version containing demo tracks and a "Making Of" bonus disc. Concept 2: Turn Up the Dial - Mac DeMarco is the face of an Apple Music rock playlist takeover event. - For one week, Mac guest DJs and curates the prominent "Rock Out" playlist on Apple Music, filling it with his favorite new rock tracks. - Playlist cover art and description changed to Mac DeMarco artwork and quotes. - Mac promotes the takeover on his social media and creates Apple Music original content like commentary tracks discussing his curation choices. - Apple Music Rock Station centered around Mac for the week - interviews, performances, influencer collaborations. - Takeover culminates in an Apple Music livestream concert performance by Mac DeMarco. Concept 3: Off the Grid - Mac DeMarco headlines an intimate, invite-only concert powered completely by Apple technology and renewable energy. - Outdoor concert staged at scenic natural location and powered by solar panels and batteries from Apple's product lineup. - All equipment used is Apple - sound, lighting, filming, etc. Showcases innovation powering creativity. - Apple promotes event with product placement in concert livestream and recap videos. Short commercial cut with voiceover tying together nature, tech, and art. - Limited merchandise capsule designed by Mac and sold by Apple. - Social media sweepstakes for Apple/Mac fans to win VIP access to the off-the-grid concert experience. Let me know if you would like me to expand on any of these concepts further!
Our next step is going to refine this output to ultimately suggest some key improvements. Let's see what we've got!
Step 2: Critiquing our Thought Starters
Here is what Claude produced as the second step of our AI App, using the prompt and context we set for it.
Here is a detailed critique of the thought starter concepts within the context of a Mac DeMarco x Apple brand partnership:
Overall Critique:
The concepts generally align with the campaign objective of showcasing the synergy between Mac DeMarco's music and Apple's technology. They highlight Apple devices as enablers of creativity in Mac's process. However, the concepts could be refined to more tightly integrate the "Simplicity in Creation" theme and focus on how the technology simplifies, rather than just enables, the creative process.
There is an opportunity to expand beyond music creation into content creation/sharing and audience engagement tactics. This would strengthen the audience targeting of tech-savvy music enthusiasts who appreciate both innovation and indie music. Collaborative fan content creation via social media or in-store experiences could achieve this.
From a superfan perspective, the concepts provide exciting behind-the-scenes access to Mac's creative process with Apple. However, the devil's advocate may argue the concepts come across as overly promotional of Apple without enough balance of Mac's artistry shining through. Ensuring both brands receive equitable spotlighting will be important.
Individual Critiques:
Concept 1: The Studio Session
This concept provides an immersive view into Mac's creative approach through a documentary format. Seeing the full album creation process first-hand would be intriguing for superfans. The focus on Apple's enabling technologies is clear.
However, the concept could go farther to emphasize simplicity and ease of use in the creative process with more specific examples. There is also room to expand the distribution and audience engagement tactics beyond just a video release.
Concept 2: Turn Up the Dial
This concept allows Mac to share his artistry through curation on a mass platform like Apple Music. Takeovers are proven engaging formats, and the livestream concert provides a climactic moment.
But it lacks a tight integration with the "Simplicity in Creation" theme. Tactics that involve Mac using Apple technology to create content, not just curate it, would be stronger. The concept also leans promotional without enough artistry.
Concept 3: Off the Grid
The intimacy and connection of an exclusive live show aligns well with Mac's vibe. The renewable energy and natural location integration is unique. It provides a fresh way to link technology, sustainability, and creativity.
However, the concept needs to focus more on the simplicity angle. Showcasing how the Apple technology integrates seamlessly to power the off-the-grid experience would strengthen the messaging. There is also room to expand the concept with collaborative fan content creation.
In summary, the thought starters present exciting possibilities but need refinement to fully achieve the potential of a Mac DeMarco x Apple partnership. Let me know if you would like me to expand on any part of this critique.This is solid feedback, and if a colleague gave this to you it would definitely help you improve your ideas.
Let's review our final step to see how Claude combines Step 1 + 2 outputs into an improved set of thought strarters.
Step 3: Creating an improved set of Thought Starters
Here is what Claude produced as the final step of our AI App, using the prompt and context we set for it.
Here is the improved and expanded set of concepts for the Mac DeMarco x Apple brand partnership campaign "Unleash Creativity":
## Executive Summary
The following document outlines three refined concepts for the "Unleash Creativity" campaign between musician Mac DeMarco and Apple. The concepts focus on showcasing Apple technology as a simplifier of the creative process for artists, using Mac DeMarco as an example.
Tactics provide an inside look at Mac's artistic approach and engage fans through collaborative content creation. The concepts aim to highlight the synergy between innovation and artistry by integrating Apple devices into Mac's signature lo-fi, indie aesthetic. Strategic considerations analyze campaign objectives, target audience appeal, and partnership equity.
## Concept 1: Simple Studio Sessions
### Overview
Mac DeMarco creates a 3-track EP exclusively using Apple products and a simplified mobile studio setup in unique locations. Apple documents the creation process through a YouTube miniseries.
### Execution
- Mac writes, records, and produces a 3-track EP titled "Mac on Apple" using only a MacBook Pro, iPad, iPhone, and select peripherals/software.
- The creative process is captured in a 4-part YouTube miniseries called "Simple Studio Sessions" airing on Apple's channel.
- Episode 1 shows Mac writing/demoing songs using Garageband on iPad and iPhone voice memos. Highlights portability and intuitive UX.
- Episode 2 captures Mac recording instruments and vocals with a simplified studio setup using MacBook Pro and peripherals. Showcases seamless ecosystem and power.
- Episode 3 shows Mac mixing/editing the tracks with Garageband on MacBook Pro. Focuses on ease-of-use in production.
- Episode 4 reveals the final EP and reflects on the simplified creative process with Apple products.
- Mac and Apple amplify "Simple Studio Sessions" through social media. Apple runs condensed versions as pre-roll/mid-roll video ads.
- The "Mac on Apple" EP is available exclusively on Apple Music first for one month.
### Tactical Considerations
- Produce high-quality documentary-style visuals to appeal to Mac's artistic fans. Use subtle product integration.
- Promote series with video trailers and social media teasers building anticipation.
- Time EP exclusivity window to align with holiday shopping season.
- Leverage influencer marketing by having Mac collaborate with prominent music/tech YouTubers on the series.
### Strategic Considerations
- Tightly aligns with "Simplicity in Creation" theme by showing simplified music production process with Apple ecosystem.
- Provides superfans an inside look at Mac's creative style in an authentic, on-brand way.
- Lengthy YouTube video series allows robust storytelling and product integration.
- Exclusive EP drives Apple Music engagement and showcases creative output enabled by Apple technology.
## Concept 2: Co-Create with Mac
### Overview
Mac engages fans by co-creating artistic content for social media using Apple products and featuring fan submissions.
### Execution
- Mac solicits 10-second music clip submissions from fans via social media to be used as samples for track creation.
- He creates a short original track using his favorite fan samples and Apple apps on an Instagram live session.
- The co-created track is posted natively on Apple Music, with a snippet on Instagram/TikTok.
- Mac hosts a live Q&A on Twitter Spaces powered by Apple Music discussing the creative process and interacting with co-creators.
- An Apple Music playlist of Mac's favorite fan-submitted samples is released.
- Apple runs social media and pre-roll ads highlighting the co-creation.
### Tactical Considerations
- Leverage UGC and fan excitement around engaging directly with Mac.
- Curate top fan sample submissions to highlight diversity.
- Time Twitter Spaces event for optimal engagement across time zones.
- Promote campaign with a branded hashtag like #CreateWithMac.
### Strategic Considerations
- Provides interactive, creative experience for fans.
- Showcases simplicity of music creation through accessible mobile apps.
- Authentic connection between Mac and his audience.
- Blends artistry, innovation, and audience participation.
- Drives organic social engagement.
## Concept 3: In-Store Album Launch
### Overview
Mac DeMarco hosts album listening sessions and pop-up performances at Apple stores to launch his new album created with Apple products and software.
### Execution
- Mac's upcoming album is teased as being created start-to-finish with Apple devices and apps.
- Launch week features "Mac @ Apple" in-store sessions in 20 flagship Apple locations globally for hands-on demos.
- Fans can test out Garageband features, accessories, and devices used by Mac.
- Photo opportunities with Mac-themed Apple store branding.
- Surprise acoustic performances by Mac DeMarco.
- Apple runs a targeted social media campaign driving to the in-store experience.
- Stores offer limited-edition co-branded Mac DeMarco x Apple merchandise.
### Tactical Considerations
- Time with upcoming album release and align with major holiday shopping season.
- Select Apple store locations based on music scene appeal.
- Limit merchandise quantity to drive urgency and exclusivity.
### Strategic Considerations
- Provides hands-on, interactive experience for superfans.
- Retail partnership boosts in-store engagement and aligns with Apple's creative positioning.
- Showcases Apple ecosystem and simplicity in an authentic way.
- Merchandise sales opportunity and media buzz.
In summary, these three refined concepts aim to authentically highlight Apple's role in simplifying creativity for artists like Mac DeMarco. Let me know if you would like me to expand further on any specific concept or strategic consideration.Wow. Talk about an improvement!
By layering the different cycles of thought into different prompts, we're able to achieve a much better output with Claude.
I also hope that after this tutorial you can see how to expand on this for your own needs.
Want to dig deeper into superfan ideas?
Create a Step 4 with a prompt that extracts the superfan-oriented ideas from Step 3 and builds on those specifically.
Want to ensure the AI stays on brand with its ideas?
Add your brand guidelines as Static Text, and then inject it into prompts alongside your partnership brief.
Still want to chat with the AI once it has come up the ideas?
Add a Chatbot widget and pass in User Input, Text Generation, and Static Text as the context.
Then you can chat with a bot that has been primed with all the context from your ideation session!
Conclusion
When used correctly: AI empowers creatives & strategists with new tools and mindsets, yielding solutions that otherwise might have been missed, all with minimal effort from the user.
You might find yourself dreaming that AI could help you with your job from end-to-end. And one day soon, that will likely be the reality.
But it's a daunting prospect for teams that currently have little to no AI augmentation in their workflow.
Instead of trying to eat the apple in one bite, approach the work bite by bite and perfect each step as you go along.
Building simple AI apps for bite-sized portions of your work process are a great place to start. :)
Let me know in the comments if you enjoyed this or found it helpful, and please follow me on LinkedIn for more content like this.
Cheers!
