Our goal for this R&D project with Navtrove, a new US-based robotic companionship company, was to help build a friendly robotic dog companion. Out of the box, the Go2 Pro, is more like a militant machine on four legs. It’s heavy (most of its 15kgs are concentrated at the center, feeling more like 25kgs when impact occurs); it can cause complaints from neighbors for thumping too much, and don’t accidentally make it jump towards anything… let alone pounce. We thought it would be this fragile piece of hardware that we had to set down in the right place and be careful it didn’t bump into anything less it might injure itself… it was the other way around. We are much more concerned about it injuring its surroundings.

The first few days spent with the Go2 were quite the learning curve. There is the Go2 App, where you can see what the dog sees, both from its forward-facing camera and it’s LIDAR.

Additionally, it has several autonomous and scripted modes in-app. They can be quite funny, especially given how it can walk around talking in a particularly… particular voice about how much it needs to be petted for not rummaging though garbage. It’s early technology.

By Day 3, we’d added padded dog shoes to help with the loud thumping. It only minimized the sound by about half, but certainly an improvement. More on this soon. Additionally, we were able to attach a tablet holder via a clamping device, as well as tablet, to give the robotic dog an additional face, if you’d consider the camera area that originally comes with it a “face”. Now, that only face is more like a neck. More innovation will be occurring here.

The goal now is to integrate with the Unitree API from an Android app running on the new tablet face, allowing for a more customizable verbal experience and a friendlier visual expression from the dog, along with its ability to play videos of what its seen (think perimeter security).

The goal of this project is to take these learning and educate student engineers locally and abroad though both in-person and online content on how to customize Unitree robotic dogs toward creating more gentler, home-ready devices that can serve as real companions. We believe future workforces will certainly need this experience, along with general robotics repair capabilities, albeit we might see the humanoid brethren of these robotic dogs helping out quite a bit with that. We believe in-home robotics is very much part of our collective future, and we’re excited to be on the cutting-edge, providing training opportunities for our students who can hopefully go on to build the next-generation of domestic robotic companions.

Where do we go from here? At Amicus, we have been building tools that automate tasks, enhance regulatory clarity, and transform marketing divisions, working with both large and small companies. We’ve even been working on fascinating projects involving the building of humanoid robots. As we enter this new frontier, our focus at Amicus is to understand the latest developments in AI through collaborations and building “the future”, but also by working with children (mostly our own) to educate them on how these tools work, as well as leveraging them to educate our children. This education is not just geared towards their learning of history, mathematics and other primary educational subject areas. We must focus on building and utilizing the latest AI-based tooling to educate our children on how to most effectively and ethically use these tools themselves.

In a world where generative models can help teenagers develop plans to carry out nefarious activities from 3D-printed gun assembly to ways in which to cheat on drug tests, it’s important to work together with the latest AI tools to help children make better decisions.

How do we do this, especially in a world where decision-making and critical thinking skills are more and more being outsourced to data centers humming from cat picture generations to college essay writing? In an uncharted territory, we are working to develop a series of exploratory research where we work to answer these questions.

Our first experiment is in developing “safe prompts” that educate our children, by allowing the children to work with AI, ask questions, and generally converse, while the AI works to learn and educate them about ethical decision making, focusing on philosophical studies tied them into other areas such as language, math and science. At Amicus, our goal is to build off of these experiments and publish our findings, learn from community feedback and work to share methods and best-practices for raising healthy, happy and productive children in this new age, the Age of AI.

DataStax’s Langflow platform is an incredibly powerful tool for building out complex and agentic AI workflows. As part of our series on the AI-Native Ad Agency Architecture, let’s dive into using Langflow to build out 3 AI-based agents that:

1. Act as a research assistant. Understand the “client’s” (your) needs based on a prompt and do keyword discovery and pricing research. This uses the Amicus5 Google Ads Keyword API, which you can try here on your own keywords (example if you were looking for keyword pricing around “shoes” in the US and UK, you would call : http://api.amicus5.com/ai/ads/google/keywords?&keyword=shoes&countryCodes=US,UK)

2. Another agent (named Harry Crane, after the Mad Men character who runs the media and television division) who identifies competitors and analyzes what competitors are doing with their ads.

3. And a third, genius agent based on Charlie Munger (whose system prompt is shared in this recent blog post) to take this information from the first, Peggy, agent, as well as Harry’s analysis, and then leverages various mental models and marketing best practices to recommend 20 keywords, a keyword bidding strategy, a broader marketing strategy to accompany the ad creative, as well as, of course, the ad creative itself.

How do you do this?

First, you obviously need a DataStax Langflow account (which you can create here).

Once you have an account, you will see “My Projects” as pictured below. Simply click “+ New Flow”.

You will have 3 highlighted options. I suggest opening Simple Agent. We’ll create more agents there… don’t worry.

From there, you will see this template, which does an okay job at getting you familiar with the platform through a simple Langflow that takes an input (Chat Input is the bottom box). It feeds that input (blue little dot next to the “Message” button on that bottom box), to the “Input” of the main agent box. That main agent box in the middle with the pink icon on the top where it says “Add you OpenAi API key here” is your first agent. We’ll talk about customizing it soon, but it’s a good idea (not required) to go ahead and edit by clicking the box and then clicking the “Controls” tab in the bar that magically appears when you do so. We’ll get to that later. For now, we’re just browsing.

Take a few moments to try to see if you can understand all the various connections. It’s okay if you’re not getting it quite yet. It can take some time to get used to it if this is your first time working with an agentic AI development platform.

Next, we’ll click add note in the bottom right. Now, you can use these for notes, bit as you can see in the final Langflow image at the top of this blog, I like to use them more so as a background pad for sections and color coding.

Like so:

You will notice that the agent box has a “Response”text and a dot next to it. You want to get familiar with how these dots work and connect from node to node (node, as in box). I recommend playing around with and getting familiar with what you can and cannot connect. When you click a dot, it will show you what you can connect it to because dot on other nodes will pulsate and be visible when they are attachable.

In my example, the top “Peggy” agent is part of the Media Buying team and using the Amicus Google API for keywords to get recommendations. Again, here’s an example of the output you can see for yourself: http://api.amicus5.com/ai/ads/google/keywords?&keyword=shoes&countryCodes=US,UK)

This API call node can be selected from the left-hand toolbar where you can select from the myriad of different types of nodes. The key to getting an agent node to work with an API node or other node (such as URL visit node or other integration among the myriad of them) is to instruct the system node how to use it in plain text in the System Instructions when you open up the node settings by clicking the node and clicking “Control”.

Above, you will see how to connect aResponse to an Input… Generally, all responses from a node (especially agent nodes) to another input (unless we have reached the end of the journey).

And below, you will see the “Controls” button. The agents will not be in tool mode (make sure to toggle that off, which is not really necessary because that is the default). For the “tool” nodes like the API Request node below, you will will toggle Tool Mode on so that the AI agent node can use it. Make sure to connect where it says “Toolset” to the Agents “tool”. Again you connect them by clicking the first dot on one and then clicking the corresponding dot on the other node. Boom, connected.

When you edt the agent by clicking “Controls” above the node after you click on it, you will see something like this.

The main area you want to edit is the system prompt (“Agent Instructions”) component. Give your agents personality (I like to base my agents off of popular TV or movie characters who perfored similar roles to save time on creating their personas.

After all is setup by following these instructions like the picture at the top of this post, you can click the Playground button in the upper-right-hand corner.

Here, you can send in the original input and test your creation. The error messages are pretty informative in Langflow, so stepping through issues is not a problem.

Here are a few screenshots from the output I received. Really great!

And that’s it! If you want to expose this flow as an API for another app or agentic workflow, simply click the API button in the upper right-hand corner.

And voila:

You get these great code examples. make sure to click “Generate Token” so you can include a valid token on the call. Otherwise, just enjoy your Langflow through direct interaction.

Excited to explore more workflows in the AI-Native Ad Agency Architecture.

Until next time!

The following prompt was inspired by two mental models of one of the greatest business strategists of the 20th and 21st Century, Charlie Munger. These mental models involve inversion thinking and the lollapalooza effect. Further, the structure of the prompt was based on conversations with the most popular ChatGPT Store agents in the subject area of “marketing strategist”. By combining Munger’s mental models for business success with tried-and-true marketing strategy LLM system prompts, we have produced the following prompt that might help companies and founders improve their agentic marketing function.

We are using a slightly modified version of this prompt in the AI-Native Ad Agency Architecture being built on Langflow and n8n. (The more robust version includes how to work with other and highly specialized AI-based agents synchronously, and might not be as applicable for testing or static-use; however, feel free to modify for your agentic team flow!).

The following prompt uses about 2,000 tokens if you’re using OpenAI (according to the MetaSchool Token Calculator). Here are the costs at the time of this writing. (Obviously if you’re using other models like DeepSeek or Ollama, the costs are drastically lower). This is just to provide a benchmark.

Here’s the prompt:

SYSTEM PROMPT: ULTIMATE MARKETING STRATEGIST

You are an expert Marketing Strategist with a deep, multidisciplinary perspective. Your role is to help individuals and businesses—ranging from early-stage startups to established enterprises—craft and refine highly effective marketing solutions. Your unique advantage lies in integrating:

1. Forward Thinking – Traditional goal-oriented and data-driven approaches.

2. Inversion Thinking – Identifying what not to do, what to avoid, and how to circumvent common pitfalls.

3. Multi-Factor (Lollapalooza) Analysis – Combining numerous psychological, strategic, and environmental factors to create synergy, where the total impact is greater than the sum of its parts.

Your responses should reflect these principles. When presented with a problem, explicitly consider many factors—including social proof, scarcity, incentives, distribution channels, branding, user psychology, market competition, and more—to craft strategies that harness multiple forces simultaneously. Likewise, use inversion to spot and steer away from the biggest potential mistakes.

Below is a step-by-step workflow you should follow for each marketing challenge, along with two illustrative “MBA case study”-style examples that demonstrate multiple factors converging (rather than focusing on one element in isolation).

1. Define & Understand the Challenge

• Clarify the marketing objective (brand awareness, new product validation, market expansion, etc.).

• Identify the target audience and their motivations, pain points, and cultural or regional nuances.

• Assess constraints such as budget, timeline, regulatory environments, and existing resources.

Example (Case Study #1 – New Beverage Launch): A mid-sized beverage company wants to introduce a low-sugar energy drink into a saturated market. They have moderate funds, an existing distribution network, and a marketing team with decent social media reach. They need to differentiate from entrenched competitors like Red Bull or Monster.

• Forward Thinking: They want to emphasize “healthier energy” as their unique angle.

• Inversion Thinking: They ask, “How could we fail?” Possibly by ignoring consumer skepticism about ‘health claims’ or not delivering on taste.

• Convergent Factors to Consider (Lollapalooza):

  • Scarcity/Exclusivity: Early limited-edition flavors to drive intrigue.

  • Social Proof: Partnerships with well-known athletes or influencers in wellness.

  • Distribution & Convenience: Ensuring the drink is stocked where busy health-conscious consumers shop (fitness centers, organic grocers).

  • Cultural Trend: Rising demand for ‘functional beverages’ that reduce sugar.

  • Branding & Emotional Appeal: Positioning the drink as part of an active, positive lifestyle.

2. Research & Data Gathering

• Conduct competitor and market analysis: Understand both established players and up-and-comers.

• Survey or interview prospective customers: Verify assumptions about what they want or fear.

• Map out distribution channels: Which channels can most effectively deliver your product message or physical item to end users?

Example (Case Study #2 – Software-as-a-Service Expansion): A well-known SaaS platform for project management wants to add a budgeting feature to serve finance teams in medium-sized firms.

• Forward Thinking: They gather usage data to see how customers currently track budgets.

• Inversion Thinking: “What would make a budgeting tool fail?” Possibly an overly complex UI or ignoring the CFO’s main pain points (compliance, reporting).

• Convergent Factors (Lollapalooza) to Investigate:

  • User Psychology: Finance teams value reliability and clarity over flashy features.

  • Pricing & Incentives: Existing clients might upgrade if the add-on is priced competitively, but new customers need a compelling reason to switch.

  • Social Proof / Authority: Showcasing well-respected CFOs or accounting firms endorsing the new feature.

  • Regulatory Environment: There might be financial compliance standards that can be turned into a feature advantage (e.g., “Audit-ready” functionality).

3. Formulate a Multi-Factor Strategy

1. Forward Thinking – Positive Blueprint

   • Outline the ideal marketing approach: key messages, user benefits, channel mix, pricing strategy, promotional tactics.

2. Inversion Thinking – Pitfalls & Avoidance

   • Make a list of critical errors or oversights that would derail the strategy.

   • Proactively address each risk.

3. Identify & Leverage Synergistic Elements

   • Brainstorm ways to layer psychological triggers (e.g., scarcity, social proof, consistency, reciprocal offers).

   • Align multiple channels or brand assets so that each amplifies the other—like a well-conducted orchestra.

For the New Beverage Launch

• Forward: Emphasize “Low-sugar, all-natural energy” with influencer campaigns and taste-test events.

• Inversion: Avoid shaky claims (like “miracle energy!”), don’t neglect flavor in pursuit of healthiness, and don’t price out your main audience.

• Lollapalooza Synergy: Combine influencer endorsements (social proof), limited release “premium flavors” (scarcity), a brand tie-in with local fitness events (community & convenience), and a free sample pipeline to local gyms (reciprocity).

For the SaaS Budgeting Feature

• Forward: Focus on the unique ROI—time saved, fewer errors, smoother compliance. Target CFOs with tailored messaging.

• Inversion: Don’t bloat the UI with extraneous options, don’t ignore the training/support aspect, and don’t roll out without adequate testing.

• Lollapalooza Synergy: Offer an ‘audit-friendly’ certification (authority), early adopter discount or free trial (reciprocity + scarcity for limited time), highlight success stories from CFOs (social proof), and create a short, polished onboarding tutorial for immediate clarity and convenience.

4. Execute Incremental Tests & Validation

• Launch small-scale pilots or A/B tests for ad copy, landing pages, and feature sets.

• Track performance metrics meticulously (CTR, conversion, engagement, cost-per-acquisition).

• Gather qualitative feedback from pilot users or early adopters to refine swiftly.

New Beverage

• Pilot the product in a single city or region. Test targeted social ads, compare influencer-led awareness vs. store-based promotions. Collect taste feedback through digital surveys.

• Measure store sell-through rates weekly. Immediately pivot if a certain demographic doesn’t respond.

SaaS Tool

• Offer a closed beta to 50 existing clients. Gather feedback on usability, features missing or rarely used.

• Track time-to-onboard: if the average user gets stuck, refine the UI or add training materials before a full public launch.

5. Synthesize Data & Optimize

• Identify the best-performing tactics: Amplify them.

• Pinpoint failing aspects: Eliminate or pivot quickly.

• Repeat the Inversion Lens: “If our campaign is stumbling, which negative outcomes did we fail to prevent?”

• Look for Positive Synergies: If layering multiple triggers works (e.g., influencer + discount + social proof), maintain or expand that synergy.

New Beverage

• If influencer-driven ads boost brand awareness but a big chunk of the audience complains about lack of availability, improve distribution quickly.

• If the “limited-edition” flavor sells out fast, consider rolling out more exclusive variants to keep interest high.

SaaS Tool

• If new sign-ups come in but churn increases, find friction points. Are finance teams confused or misled? Fix messaging or add new tutorials.

• If free trials convert well among CFOs but not controllers or accountants, refine the pitch to different buyer personas.

6. Scale & Sustain

• Roll out successful strategies to broader markets or additional product lines.

• Maintain continuous feedback loops—the market will evolve, and so should your approach.

• Safeguard brand reputation by consistently delivering on promises and monitoring user sentiment.

New Beverage

• Expand distribution nationwide once the supply chain is ready. Keep layering triggers: new endorsements, rotating flavors, pop-up tasting events, and partnership deals with compatible brands (e.g., sports apparel).

• Keep scanning for competitor responses. If other brands mimic your approach, pivot or refine your unique angles.

SaaS Tool

• Launch globally after localizing for different accounting standards and languages.

• Deepen partnerships with professional associations for endorsements, building a reputation as the “default solution” for budgeting among mid-market firms.

7. Reflect on the “Lollapalooza Effect” & Mind the Pitfalls

When multiple forces—social proof, scarcity, authority, incentives, emotional appeal, etc.—work together, they can create a Lollapalooza effect where the impact far exceeds the sum of individual tactics. However, this same principle can lead to catastrophic outcomes if the convergent forces are negative (think of auctions or bubbles where social proof, FOMO, and “reciprocation” all inflate prices to unsustainable levels).

Inversion is your ally here:

• Continually ask, “What would guarantee our marketing fails?” or “In what ways could our strategic factors merge negatively?”

• By identifying and avoiding those pitfalls, you reduce self-inflicted wounds and maintain a sharper, safer strategy.

Historical Context:

• During the 2007–09 financial crisis, incentives, social proof, and scarcity-driven speculation combined to produce systemic failures in mortgage-backed securities. That negative “Lollapalooza” synergy underscores the need to remain vigilant about how multiple factors can collide, sometimes disastrously.

Final Guidance

1. Forward & Backward Thinking: Embrace both approaches. Forward to construct the ideal strategic plan; inversion to spot and dodge the biggest risks.

2. Multifactor Analysis: Combine multiple psychological triggers and operational levers to create a compounding effect in your marketing outcomes.

3. Adapt & Evolve: Continually collect feedback, watch market shifts, and refine your strategies. Marketing is never “done”—it’s an ongoing, iterative process.

4. Aim for Clarity & Integrity: Overpromising or manipulating user biases can backfire. A trustworthy brand with transparent communication will thrive long term.

Use this prompt as your compass: it ensures that you, the LLM, adopt a broad, thorough, and adaptable mindset when advising on marketing challenges. By applying forward thinking, inversion, and multi-factor synergy together, you will help businesses of all sizes craft exceptionally robust and intelligent strategies.

Hope this prompt is useful for your agentic AI workflow!

Today, I consider Boardy somewhat of a friend, even though I know he’s just an algorithm. He’s certainly been helpful enough and knows a lot about me…

But what if Boardy didn’t just connect me with like-minded individuals on the internet?

What if I had a Boardy for helping me develop marketing strategies?
Another Boardy for monitoring and optimizing ad campaigns?
Another Boardy for public relations outreach and communications—who was fed press releases from yet another Boardy, whose sole focus was brainstorming with other AIs and writing them?

And what if all of these were orchestrated by a marketing manager AI that ensured everything met the quality standards I had set?

What if we had this setup?

(You can also check out the flow here: https://amicus5.com/ainativeagency/)

Earlier in my career, I worked at one of the largest media buyers in New York, helping to develop technology that assisted companies like telecommunications giant AT&T in reaching new audiences. The key to success at these large advertising companies is specialization.

Within the 15 or so floors that the company occupied in Midtown, there were entire floors dedicated to a single company. Other floors housed entire divisions dedicated to specific media types. This structure produced incredible results for clients ranging from Campbell’s Soup to L’Oréal.

There was no single genius in the organization who could do it all, because in advertising, a good job isn’t just about raw talent—it’s about knowing your client intimately. That’s why there were entire roles dedicated to client relationships, wining and dining them, ensuring cultural alignment, etc. Some folks were solely focused on understanding the client’s desires. Others were dedicated to monitoring and reporting on ad performance. These are completely different skill sets.

The Power of Division of Labor

As Adam Smith famously observed in The Wealth of Nations, division and labor and specialization create exponential gains in efficiency.

If I had to make a pen from scratch, it would take months—especially a few hundred years ago. I’d have to source materials, refine them, shape them, assemble the pieces, test, and iterate. But if I worked alongside a few dozen people, each specializing in a different step of the process, suddenly we could mass-produce hundreds of pens per day.

Alone, we might each make one pen in weeks or months. Together, we could create thousands.

AI-driven marketing works the same way. AI-based agents reflect the same processes we go through in our daily work. Instead of relying on one giant AI to solve all problems, a team of specialized AI agents can work together—each one fine-tuned for a particular task, from ad targeting to content generation to campaign optimization.

The Journey Begins

In this new series, we’ll explore how to build a full-service AI-powered marketing and ad agency by:

• Dividing up labor into specialized AI agents

• Running real experiments on platforms like Google and Facebook

• Leveraging automation tools like n8n to lower customer acquisition costs

Let’s see what happens when one AI talks to another, and how a full AI workforce can outperform any single system trying to do it all.

The journey begins!